{"family":"models-ops","name":"Models, Serving & Ops","count":134,"license":"CC-BY-4.0","systems":[{"id":"01-ai-yi-family--01-ai","name":"01.AI Yi family","url":"https://www.01.ai/","section":"Foundation models (substrate reference)","tier":2,"status":"active","last_verified_at":"2026-10-31","type":"Open-weights frontier-tier Chinese model family (Yi-Lightning / Yi-34B)","license":"Apache 2.0 (Yi-34B + base sizes); proprietary (Yi-Large, Yi-Lightning)","latest_release":"Yi-Lightning (2024-10)","deployment":"Self-hostable (Yi-34B Apache 2.0) + 01.AI API for Yi-Large / Yi-Lightning; Together AI hosts Yi-34B","pricing":"API: Yi-Lightning $0.14 per 1M tokens (vendor-claimed lowest among frontier-tier); Yi-34B weights free under Apache 2.0 (international API at platform.01.ai suspended; API access now only via platform.lingyiwanwu.com)","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"open (Apache 2.0 for Yi-34B); single-vendor for Yi-Lightning","github":"github.com/01-ai/Yi ~7k+ stars","description":"Kai-Fu Lee's 01.AI (founded 2023, Beijing). Yi family open-weights — Yi-34B (2023), Yi-Large (proprietary 2024), Yi-Lightning (2024-10, frontier-tier, matched GPT-4o on LMSYS chatbot arena). Apache 2.0 (Yi-34B and base sizes). Important second Chinese open-weights option alongside DeepSeek + Qwen. By 2026 01.AI's site leads with enterprise AI products (WorldWise Enterprise LLM Platform, 2025-03; Boss AI; TopSales AI) and lists Yi-Lightning (2024-10) as its latest Yi model. 01.AI's international API platform (platform.01.ai) is suspended: it says API services are temporarily unavailable from August 25 due to business adjustments and points users to the Chinese platform at platform.lingyiwanwu.com.","cited_cells":34,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/01-ai-yi-family--01-ai"},{"id":"1x-technologies--1x-tech","name":"1X Technologies","url":"https://www.1x.tech/","section":"Robotics foundation models & agent stacks","tier":2,"status":"active","last_verified_at":"2026-10-26","type":"Humanoid robot maker (Neo / Eve) — OpenAI portfolio","license":"","latest_release":"NEO home humanoid (orders opened 2025-10-28; US deliveries slated 2026); NEO hands platform (2026-07-09)","deployment":"Direct customer (industrial Eve + consumer Neo)","pricing":"NEO: $20,000 early-access purchase (3-yr warranty) or $499/month subscription; $200 refundable deposit","mcp_support":"","a2a_support":"","backend_storage":"Internal 1X cloud","llm_lock":"","github":"","description":"Norway / US humanoid robot company — Eve (wheeled bimanual) + Neo (legged humanoid, consumer-targeted, late 2025 / 2026 launch). Raised $100M Series B (Jan-2024, EQT + OpenAI Startup Fund); $40M Series A (Mar-2023, OpenAI lead). Distinguishing feature: consumer-home positioning rather than industrial.","cited_cells":33,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/1x-technologies--1x-tech"},{"id":"accelerate-hugging-face--gh-huggingface-accelerate","name":"Accelerate (Hugging Face)","url":"https://github.com/huggingface/accelerate","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-26","type":"Distributed training abstraction","license":"Apache-2.0","latest_release":"v1.15.0 (2026-09-09)","deployment":"Self-hosted only — pip-installable Python library (pip install accelerate); runs on local, multi-GPU, TPU and multi-node clusters","pricing":"Free and open source (Apache-2.0); no paid tier","mcp_support":"no","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"huggingface/accelerate — ~9.9k stars, last commit 2026-09 (latest release tag)","description":"Hugging Face's abstraction over distributed-training backends (DDP, FSDP, DeepSpeed, Megatron). Minimal code change to scale a PyTorch script.","cited_cells":20,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/accelerate-hugging-face--gh-huggingface-accelerate"},{"id":"agent-router-formerly-envoy-ai-gateway--theagentrouter-ai","name":"Agent Router (formerly Envoy AI Gateway)","url":"https://theagentrouter.ai/","section":"Inference platforms & gateways","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"Apache-2.0","latest_release":"last commit 2026-10-28","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/theagentrouter/agent-router (2314★)","description":"Open-source, Envoy-powered control plane giving agents one OpenAI-compatible API for every model and one router for MCP tools, with provider credentials, failover and token limits. It was formerly Envoy AI Gateway and is now an Agentic AI Foundation project.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agent-router-formerly-envoy-ai-gateway--theagentrouter-ai"},{"id":"agentgateway--agentgateway-dev","name":"agentgateway","url":"https://agentgateway.dev/","section":"Inference platforms & gateways","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"Apache-2.0","latest_release":"last commit 2026-10-28","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/agentgateway/agentgateway (5769★)","description":"Open-source Rust proxy for AI agents covering LLM routing with credentials and failover, native MCP and A2A traffic, and token-usage limits. Contributors include Solo.io, Microsoft and AWS, and kgateway uses it as its AI data plane.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agentgateway--agentgateway-dev"},{"id":"agility-robotics--agilityrobotics-com","name":"Agility Robotics","url":"https://agilityrobotics.com/","section":"Robotics foundation models & agent stacks","tier":2,"status":"active","last_verified_at":"2026-10-26","type":"Bipedal warehouse robot maker (Digit) — first US humanoid commercial deployment","license":"","latest_release":"Digit 5 (unveiled 2026-09-15; early access H1 2027, GA by end-2027)","deployment":"Warehouse deployment; RaaS model","pricing":"Robot-as-a-Service (RaaS) — vendor reports per-hour rates","mcp_support":"","a2a_support":"","backend_storage":"Internal cloud","llm_lock":"","github":"","description":"Oregon-based humanoid maker; Digit is in production at GXO Logistics warehouses since Sep-2024 — claimed first humanoid in commercial service in the US. Spinout of Oregon State University. $400M+ raised (Amazon Industrial Innovation Fund participant).","cited_cells":35,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agility-robotics--agilityrobotics-com"},{"id":"alibaba-qwen-3-family--qwenlm-github-io","name":"Alibaba Qwen 3 family","url":"https://qwenlm.github.io/blog/qwen3/","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Open-weights frontier model family (Qwen 3 dense + MoE; 0.5B–235B)","license":"Apache 2.0 (most Qwen 3 sizes); some larger MoE variants under Qwen License","latest_release":"Qwen3.8 open weights: flagship Qwen3.8-2.4T-A95B + Qwen3.8-27B (Apache-2.0) (2026-08-17)","deployment":"Self-hostable (Apache 2.0 weights) + Alibaba Cloud Model Studio + Together AI + Fireworks + HuggingFace inference + Ollama","pricing":"Free open-weights (Apache 2.0); Alibaba Cloud API pay-per-token ($0.10-$2 per 1M depending on tier); Together / Fireworks hosted","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"open (Apache 2.0)","github":"github.com/QwenLM/Qwen3 ~30k+ stars; cumulative QwenLM org repos ~80k+ stars","description":"Alibaba Cloud's Qwen 3 family (2025-04) — open-weights frontier models from 0.5B to 235B-A22B MoE. Apache 2.0 (most sizes) — broadest open-weights family. Includes Qwen 3 32B dense, Qwen 3 30B-A3B MoE, Qwen 3 235B-A22B MoE. Qwen-Coder, Qwen-VL (vision), Qwen-Audio companion variants. Dominant base model for many Chinese AI startups and a top OSS substrate globally.","cited_cells":53,"third_party_cells":16,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/alibaba-qwen-3-family--qwenlm-github-io"},{"id":"amazon-bedrock-guardrails--aws-amazon-com","name":"Amazon Bedrock Guardrails","url":"https://aws.amazon.com/bedrock/guardrails/","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Configurable safeguards that filter harmful text and image content, redact sensitive information, block denied topics and detect hallucinations. They apply consistently across foundation models in Amazon Bedrock and self-hosted or third-party models.","cited_cells":9,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/amazon-bedrock-guardrails--aws-amazon-com"},{"id":"amazon-nova-family--aboutamazon-com","name":"Amazon Nova family","url":"https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws","section":"Foundation models (substrate reference)","tier":2,"status":"active","last_verified_at":"2026-10-31","type":"AWS-native foundation model family (Nova Pro / Lite / Micro / Premier / Canvas / Reel)","license":"","latest_release":"Nova 2.5 Sonic speech-to-speech (2026-10); Nova 2 Lite reasoning model (2025-12-02) is newest text model on Bedrock","deployment":"Managed-only — Amazon Bedrock (custom Nova models also deployable via SageMaker Inference); weights closed","pricing":"AWS Bedrock pay-per-token: Nova Micro $0.035/$0.14 per 1M; Lite $0.06/$0.24; Pro $0.80/$3.20; Premier $2.50/$12.50","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"single-vendor (AWS Bedrock-only)","github":"","description":"Amazon's first-party foundation model family, launched at AWS re:Invent Dec-2024. Nova Pro (multimodal mid-tier), Nova Lite (multimodal cheap), Nova Micro (text-only fast), Nova Canvas (image gen), Nova Reel (video gen). Premier frontier-tier launched 2025. Substrate for AWS Bedrock-native agents and Amazon Q. Distinct from Claude / Llama / Mistral hosted via Bedrock — Nova is Amazon's own training.","cited_cells":32,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/amazon-nova-family--aboutamazon-com"},{"id":"anthropic-claude-foundation-models--anthropic-com","name":"Anthropic Claude (foundation models)","url":"https://www.anthropic.com/claude","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Frontier foundation model family (Sonnet 4 / Opus 4.5 / Haiku 4.5)","license":"","latest_release":"Claude Opus 5.5 (2026-09-22); Claude Fable 5.1 (2026-09-01); Sonnet 5.5 + Haiku 5.5 current","deployment":"Managed-only — Claude API/Platform plus Amazon Web Services, Google Cloud and Microsoft Azure","pricing":"API per 1M in/out: Fable 5.1 $10/$50; Opus 5.5 $4/$20; Sonnet 5.5 $2/$10; Haiku 5.5 from $0.10/$0.50","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"single-vendor (Anthropic-only family)","github":"","description":"Claude family of frontier foundation models — Sonnet 4 (workhorse), Opus 4.5 (frontier capability), Haiku 4.5 (fast/cheap). Frontier model used as substrate by Claude Code, Anthropic Memory tool, Cline, Aider, Continue.dev, Cursor, Goose, and dozens of other catalog entries. Memory in Claude is parametric (in weights) — discrete episodic memory lives outside the model in the Claude Memory tool (see Platform-provider memory section).","cited_cells":57,"third_party_cells":10,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/anthropic-claude-foundation-models--anthropic-com"},{"id":"anyscale--anyscale-com","name":"Anyscale","url":"https://www.anyscale.com/","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Ray-based AI platform","license":"proprietary; Ray Apache-2.0","latest_release":"v0.26.112 (2026-10-20) — anyscale","deployment":"Both (self-hosted + managed cloud) — Anyscale-hosted, or BYOC in your own cloud/VPC/on-prem (VMs or Kubernetes)","pricing":"Pay-as-you-go compute billed in Anyscale credits per hour (e.g. A100 ~4.96 AC/hr); $100 free credit for new users; committed contracts with volume discounts via sales","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"ray-project/ray — ~44.1k stars, last commit 2026-10","description":"Commercial platform built on Ray — distributed training, fine-tuning, serving. Anyscale Endpoints provides hosted OSS inference; ray cluster is the substrate underneath OpenAI's training stack.","cited_cells":21,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/anyscale--anyscale-com"},{"id":"apptronik--apptronik-com","name":"Apptronik","url":"https://apptronik.com/","section":"Robotics foundation models & agent stacks","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Humanoid robot maker (Apollo) — Google DeepMind partner","license":"","latest_release":"Apollo (humanoid)","deployment":"Direct customer (manufacturing + logistics)","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"Internal Apptronik cloud","llm_lock":"","github":"","description":"Apollo humanoid maker, spun out of University of Texas Human Centered Robotics Lab. Raised $350M Series A (Feb-2025; B Capital + Capital Factory); Mercedes-Benz manufacturing partner (Mar 2024). Partnered with Google DeepMind on Gemini Robotics (Mar 2025) — Apollo is the headline reference platform for Gemini Robotics.","cited_cells":32,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/apptronik--apptronik-com"},{"id":"argilla--argilla-io","name":"Argilla","url":"https://argilla.io/","section":"Training infrastructure","tier":2,"status":"acquired","last_verified_at":"2026-10-29","type":"Data-quality + labelling for LLMs","license":"Apache-2.0","latest_release":"v2.8.0 (2025-03-10)","deployment":"Self-hosted only — server deployed via Hugging Face Spaces template; Python SDK client","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"argilla-io/argilla — ~5.2k stars, last commit 2025-08","description":"Open-source data-quality platform for LLM training data — human-in-the-loop labelling, RLHF data collection, dataset curation. Acquired by Hugging Face 2024.","cited_cells":19,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/argilla--argilla-io"},{"id":"arize-ai--arize-com","name":"Arize AI","url":"https://arize.com/","section":"Evaluation & observability platforms","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"ML + LLM observability platform","license":"proprietary; Python SDK Apache-2.0","latest_release":"v8.61.0 (2026-10-26) — arize","deployment":"Both (self-hosted + managed cloud) — AX SaaS; self-hosted on AX Enterprise only","pricing":"AX Free $0/mo (25k spans/mo); AX Pro $50/mo (50k spans/mo); AX Enterprise custom (SaaS or self-hosted); Phoenix OSS free","mcp_support":"yes — MCP servers and skills for coding agents to set up Arize AX","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Production ML monitoring + LLM observability (Arize AX) — drift, eval, root-cause. Pre-LLM observability leader; now LLM-pivoted.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/arize-ai--arize-com"},{"id":"arize-phoenix--phoenix-arize-com","name":"Arize Phoenix","url":"https://phoenix.arize.com/","section":"Evaluation & observability platforms","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"OSS LLM tracing + eval","license":"Elastic-2.0","latest_release":"v20.26.1 (2026-10-25)","deployment":"searched not found","pricing":"searched not found","mcp_support":"yes — built-in remote MCP server at /mcp; legacy stdio @arizeai/phoenix-mcp in maintenance","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"Arize-ai/phoenix — ~12.0k stars, last commit 2026-10","description":"Arize AI's open-source LLM observability tool — OpenTelemetry-native LLM tracing, evals, datasets. Arize commercial offers Phoenix + Surveyor.","cited_cells":39,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/arize-phoenix--phoenix-arize-com"},{"id":"arthur-engine--gh-arthur-ai-arthur-engine","name":"Arthur Engine","url":"https://github.com/arthur-ai/arthur-engine","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"MIT","latest_release":"last commit 2026-10-28","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/arthur-ai/arthur-engine (93★)","description":"Open-source service from Arthur AI for enforcing guardrails in LLM and agent workflows, with real-time detection of PII leakage, hallucination, prompt injection and toxicity. It also provides trace-based evaluation and observability for agentic applications.","cited_cells":15,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/arthur-engine--gh-arthur-ai-arthur-engine"},{"id":"aws-sagemaker--aws-amazon-com","name":"AWS SageMaker","url":"https://aws.amazon.com/sagemaker/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"AWS ML platform","license":"proprietary; Python SDK Apache-2.0","latest_release":"v3.27.0 (2026-10-21) — sagemaker","deployment":"Managed-only — AWS","pricing":"Pay-as-you-go by instance type and duration, no minimum fees; free tier for new users; savings plans for commitments","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"aws/sagemaker-python-sdk — ~2.3k stars, last commit 2026-10","description":"AWS's managed ML platform — Studio, Training, HyperPod (large-scale clusters), Endpoints. Industry-standard for enterprise AWS-hosted training.","cited_cells":16,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/aws-sagemaker--aws-amazon-com"},{"id":"axolotl--gh-axolotl-ai-cloud-axolotl","name":"Axolotl","url":"https://github.com/axolotl-ai-cloud/axolotl","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"YAML-driven LLM fine-tuning","license":"Apache-2.0","latest_release":"v0.21.1 (2026-10-27)","deployment":"Self-hosted only — library/CLI with Docker images; runs on third-party GPU clouds","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"axolotl-ai-cloud/axolotl — ~12.6k stars, last commit 2026-10","description":"Open-source LLM fine-tuning library — single YAML config drives full / LoRA / QLoRA / DPO training across most popular base models.","cited_cells":19,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/axolotl--gh-axolotl-ai-cloud-axolotl"},{"id":"azure-ai-content-safety-prompt-shields--learn-microsoft-com","name":"Azure AI Content Safety Prompt Shields","url":"https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/jailbreak-detection","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Azure AI Content Safety API that detects user prompt attacks and indirect (document) prompt injection aimed at manipulating a model's behavior, alongside groundedness, protected-material and task-adherence checks.","cited_cells":10,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/azure-ai-content-safety-prompt-shields--learn-microsoft-com"},{"id":"azure-machine-learning--azure-microsoft-com","name":"Azure Machine Learning","url":"https://azure.microsoft.com/en-us/products/machine-learning","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Azure ML platform","license":"proprietary; Python SDK (azure-ai-ml) MIT","latest_release":"v1.35.2 (2026-10-12) — azure-ai-ml","deployment":"Managed-only — Azure","pricing":"No additional charge for Azure Machine Learning; pay for underlying compute and other Azure services consumed","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"searched not found","description":"Microsoft Azure's managed ML platform — pipelines, AutoML, Designer, prompt flow (LLM). Heavy integration with Azure OpenAI and Azure AI Studio.","cited_cells":13,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/azure-machine-learning--azure-microsoft-com"},{"id":"baseten--baseten-co","name":"Baseten","url":"https://www.baseten.co/","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"ML model serving cloud","license":"proprietary; Truss MIT","latest_release":"v0.18.36 (2026-10-21) — truss","deployment":"Both (self-hosted + managed cloud) — Baseten Cloud, self-hosted in your VPC (Enterprise), or hybrid","pricing":"Basic $0/mo pay-as-you-go; dedicated deployments billed per GPU-minute (e.g. H100 $0.10833/min); Model APIs per 1M tokens; Pro and Enterprise by quote","mcp_support":"yes — Baseten MCP servers and skill for agents to manage the workspace and search docs","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"basetenlabs/truss — ~1.2k stars, last commit 2026-10","description":"Inference platform for ML models — Truss package format, Chains for multi-model workflows, autoscaling GPU/CPU serving. Targeted at production teams shipping LLM/diffusion endpoints.","cited_cells":23,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/baseten--baseten-co"},{"id":"bifrost-maxim-ai--getmaxim-ai","name":"Bifrost (Maxim AI)","url":"https://www.getmaxim.ai/bifrost","section":"Inference platforms & gateways","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"Apache-2.0","latest_release":"last commit 2026-10-29","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/maximhq/bifrost (9198★)","description":"Open-source Go AI gateway from Maxim AI with one OpenAI-compatible API for many models, failover, load balancing, semantic caching, virtual keys and budgets, and a built-in MCP gateway for agents.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/bifrost-maxim-ai--getmaxim-ai"},{"id":"bonito--gh-batsresearch-bonito","name":"Bonito","url":"https://github.com/BatsResearch/bonito","section":"Training infrastructure","tier":3,"status":"stale","last_verified_at":"2026-10-27","type":"Conditional task generator for instruction tuning","license":"BSD-3-Clause","latest_release":"v0.1.0 (2025-02-12) — bonito-llm","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"BatsResearch/bonito — ~830 stars, last commit 2025-07","description":"Open model that generates instruction-tuning data conditioned on unstructured text — adapts pretrained LLMs to specialised tasks without manual labels (Brown Univ).","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/bonito--gh-batsresearch-bonito"},{"id":"braintrust--braintrust-dev","name":"Braintrust","url":"https://www.braintrust.dev/","section":"Evaluation & observability platforms","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Eval + observability for AI","license":"proprietary; SDKs open source","latest_release":"v0.48.1 (2026-10-26) — braintrust Python SDK","deployment":"Both (self-hosted + managed cloud) — hosted SaaS; on-prem deployment on Enterprise","pricing":"Starter $0/mo; Pro $249/mo; Enterprise custom","mcp_support":"yes — Braintrust MCP server (Connect with MCP quickstart)","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Eval-first LLM platform — datasets, eval runs, prompt playground, tracing. Founded by ex-Impira (acquired by Dropbox) team. Used by Notion, Brex, AirOps.","cited_cells":43,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/braintrust--braintrust-dev"},{"id":"cleanba--gh-vwxyzjn-cleanba","name":"Cleanba","url":"https://github.com/vwxyzjn/cleanba","section":"Training infrastructure","tier":4,"status":"research-code","last_verified_at":"2026-10-29","type":"Reproducible RL baselines","license":"MIT","latest_release":"no releases (no GitHub releases or PyPI package)","deployment":"Self-hosted only — research reference code","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"vwxyzjn/cleanba — ~127 stars, last commit 2024-08","description":"Distributed PPO/IMPALA reference implementations from CleanRL author — research-friendly, reproducible. README states the repo is intended for archiving purposes, with future maintenance in CleanRL; last push 2024-08.","cited_cells":13,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/cleanba--gh-vwxyzjn-cleanba"},{"id":"clearml--clear-ml","name":"ClearML","url":"https://clear.ml/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"ML experiment + orchestration","license":"Apache-2.0 (Core)","latest_release":"v2.1.12 (2026-08-19)","deployment":"Both (self-hosted + managed cloud) — hosted Community/Pro; VPC or on-prem on Scale/Enterprise; open-source server self-hostable","pricing":"Community $0 (up to 3 users); Pro $15/user/mo plus usage; Scale and Enterprise custom quote","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"clearml/clearml — ~6.9k stars, last commit 2026-10","description":"Open-source ML experiment tracking, orchestration, and data management. Allegro AI commercial; widely used in research labs.","cited_cells":19,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/clearml--clear-ml"},{"id":"cloudflare-ai-gateway--developers-cloudflare-com","name":"Cloudflare AI Gateway","url":"https://developers.cloudflare.com/ai-gateway/","section":"Inference platforms & gateways","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Gateway in front of 20+ model providers with an OpenAI-compatible unified API, caching, rate and spend limits, dynamic routing, guardrails, DLP and BYOK, plus logging and analytics.","cited_cells":13,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/cloudflare-ai-gateway--developers-cloudflare-com"},{"id":"cohere-command-r-command-a--cohere-com","name":"Cohere Command R+ / Command A","url":"https://cohere.com/command","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Enterprise-focused frontier model family (Command A / Command R+ / Command R7B)","license":"Apache-2.0 (Command A+ open weights, 2026-05); earlier Command R/A weights CC-BY-NC-4.0; API proprietary","latest_release":"Command A+ (command-a-plus-05-2026) (2026-05-20)","deployment":"Managed cloud + customer VPC (AWS / Azure / GCP / Oracle Cloud) + on-prem; weights downloadable for non-commercial via HuggingFace","pricing":"API: Command A $2.50/$10 per 1M; Command R+ $2.50/$10; Command R7B $0.0375/$0.15; Embed + Rerank separately priced","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"multi-vendor option (Cohere-hosted + customer VPC + on-prem)","github":"github.com/cohere-ai cumulative ~5k+ stars; various model repos on HuggingFace","description":"Toronto-based foundation-model lab (Aidan Gomez, ex-Google Brain 'Attention Is All You Need' co-author). Command family is enterprise / RAG-optimised: Command A (Mar-2025, 111B dense, 256k ctx), Command R+ (104B MoE), Command R7B (smallest tier). Announced merger with Aleph Alpha April-2026. Open weights via CC-BY-NC (non-commercial); Cohere-hosted for commercial.","cited_cells":38,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/cohere-command-r-command-a--cohere-com"},{"id":"collaborative-robotics--co-bot","name":"Collaborative Robotics","url":"https://www.co.bot/","section":"Robotics foundation models & agent stacks","tier":3,"status":"active","last_verified_at":"2026-10-27","type":"Collaborative robot maker (Brad Porter / ex-Amazon Robotics)","license":"","latest_release":"","deployment":"Direct customer pilot","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"Internal cloud","llm_lock":"","github":"","description":"Founded 2022 by Brad Porter (ex-Amazon Robotics VP + Scale AI CTO). Stealthier than the humanoid pack; raised $100M Series B Apr-2024 (General Catalyst). Targets warehouse + healthcare collaborative robots.","cited_cells":28,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/collaborative-robotics--co-bot"},{"id":"colossalai--gh-hpcaitech-colossalai","name":"ColossalAI","url":"https://github.com/hpcaitech/ColossalAI","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"All-in-one large-scale training","license":"Apache-2.0","latest_release":"v0.5.0 (2025-06-04)","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"hpcaitech/ColossalAI — ~41.4k stars, last commit 2026-10","description":"HPC-AI Tech's distributed training framework — ZeRO + tensor/pipeline parallel + heterogeneous training (CPU offload). HPC-AI raised $50M.","cited_cells":16,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/colossalai--gh-hpcaitech-colossalai"},{"id":"comet-ml--comet-com","name":"Comet ML","url":"https://www.comet.com/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"ML experiment tracking + LLM eval","license":"proprietary; comet-ml SDK MIT; Opik Apache-2.0","latest_release":"v3.58.8 (2026-10-08) — comet-ml","deployment":"Both (self-hosted + managed cloud) — Comet cloud; Opik self-hosted via Docker Compose/Helm; on-prem on Enterprise","pricing":"Opik: open source free, Free Cloud $0, Pro Cloud $19/mo, Enterprise custom; Experiment Management: Free, Pro $19/user/mo, Enterprise custom","mcp_support":"yes — opik-mcp MCP server for Claude Code, Cursor, VS Code","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"comet-ml/opik — ~23.0k stars, last commit 2026-10","description":"ML experiment tracking + LLM evaluation platform (Opik for LLM eval). Comet competes head-to-head with Weights & Biases.","cited_cells":21,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/comet-ml--comet-com"},{"id":"composer-mosaicml--gh-mosaicml-composer","name":"Composer (MosaicML)","url":"https://github.com/mosaicml/composer","section":"Training infrastructure","tier":2,"status":"archived","last_verified_at":"2026-10-27","type":"PyTorch training library","license":"Apache-2.0","latest_release":"v0.32.1 (2025-07-26) — mosaicml","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"mosaicml/composer — ~5.5k stars, archived 2026-10-23, last push 2026-04","description":"Open-source training library from MosaicML (Databricks) — Trainer + recipes for fast convergence; underlies the MPT model training stack.","cited_cells":17,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/composer-mosaicml--gh-mosaicml-composer"},{"id":"confident-ai-deepeval--confident-ai-com","name":"Confident AI / DeepEval","url":"https://www.confident-ai.com/","section":"Evaluation & observability platforms","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"LLM eval framework + cloud","license":"Apache-2.0 (DeepEval)","latest_release":"v4.3.1 (2026-10-16) — deepeval","deployment":"Both (self-hosted + managed cloud) — Confident AI cloud; dedicated on-prem on Enterprise; DeepEval runs locally","pricing":"Free $0; Starter $200/mo; Team $2,000/mo; Enterprise custom","mcp_support":"yes — Confident AI MCP server for Claude Code/Cursor; DeepEval MCP-use evaluation metrics","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"confident-ai/deepeval — ~19.1k stars, last commit 2026-10","description":"Open-source DeepEval framework + Confident AI cloud — pytest-style eval runs, hallucination/relevance/bias metrics, datasets.","cited_cells":23,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/confident-ai-deepeval--confident-ai-com"},{"id":"covariant--covariant-ai","name":"Covariant","url":"https://covariant.ai/","section":"Robotics foundation models & agent stacks","tier":2,"status":"active","last_verified_at":"2026-10-29","type":"Robotics foundation-model company (warehouse pick-and-place) — acqui-hired by Amazon Aug 2024","license":"","latest_release":"RFM-1 (Mar 2024)","deployment":"On-prem (warehouse) + cloud hybrid","pricing":"Enterprise; per-robot / per-pick fees","mcp_support":"","a2a_support":"","backend_storage":"Customer trajectory data on Covariant + customer infra","llm_lock":"","github":"","description":"Pieter Abbeel / Peter Chen-founded robotics FM company — built RFM-1 (Robotics Foundation Model, Mar 2024) for warehouse pick-and-place. In Aug 2024 Amazon took a non-exclusive license to its robotic foundation models and hired founders Abbeel, Chen and Rocky Duan plus research staff; Covariant was not acquired and continued serving customers. As of 2026-10-29 covariant.ai is a placeholder page (logo, video and a 2026 copyright line only; no product or contact pages).","cited_cells":39,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/covariant--covariant-ai"},{"id":"crowdstrike-falcon-guardian-formerly-pangea-ai-guard--crowdstrike-com","name":"CrowdStrike Falcon Guardian (formerly Pangea AI Guard)","url":"https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/","section":"Guardrails & AI security","tier":3,"status":"acquired","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"AI detection and response product that discovers, governs and secures AI agents at runtime across endpoint, cloud and SaaS. It is built on Pangea, which CrowdStrike agreed to acquire in September 2025, and pangea.cloud's AI Guard page now redirects here.","cited_cells":9,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/crowdstrike-falcon-guardian-formerly-pangea-ai-guard--crowdstrike-com"},{"id":"dagster--dagster-io","name":"Dagster","url":"https://dagster.io/","section":"Training infrastructure","tier":1,"status":"acquired","last_verified_at":"2026-10-29","type":"Data + ML pipeline orchestrator","license":"Apache-2.0","latest_release":"v1.13.27 (2026-10-26)","deployment":"Both (self-hosted + managed cloud) — Dagster+ Serverless or Hybrid; OSS self-hosted","pricing":"Dagster+ Solo $10/mo, Starter $100/mo (plus per-credit and serverless compute usage), Pro custom","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"dagster-io/dagster — ~16.3k stars, last commit 2026-10","description":"Data orchestrator with first-class ML pipeline support — software-defined assets model; competes with Airflow/Prefect for ML pipelines. Acquired by Prefect (announced 2026-07-13); Dagster continues as a supported product under its own name and open-source license.","cited_cells":19,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/dagster--dagster-io"},{"id":"datadog-agent-observability-llm-observability--datadoghq-com","name":"Datadog Agent Observability (LLM Observability)","url":"https://www.datadoghq.com/products/ai/agent-observability/","section":"Evaluation & observability platforms","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Datadog's product for evaluating, tracing and monitoring AI agents, combining offline experiments on golden datasets with production tracing and troubleshooting. Its former LLM Observability URL now redirects here.","cited_cells":15,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/datadog-agent-observability-llm-observability--datadoghq-com"},{"id":"deepseek-r1-v3-family--deepseek-com","name":"DeepSeek R1 / V3 family","url":"https://www.deepseek.com/","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Open-weights reasoning-first frontier model family (R1 / V3 / V3.1)","license":"MIT (R1 + V3 weights)","latest_release":"DeepSeek-V4.1-Flash-1017 (2026-10-17); DeepSeek-V4-Pro GA (2026-08-13)","deployment":"Self-hostable (MIT weights) + hosted DeepSeek API + Together / Fireworks / HuggingFace inference","pricing":"API per 1M in (cache miss)/out, peak: V4-Pro $1.32/$3.96, V4.1-Flash $0.30/$1.20; off-peak 50% lower; weights free","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"open (MIT license; many hosts)","github":"github.com/deepseek-ai/DeepSeek-V3 ~94k stars; github.com/deepseek-ai/DeepSeek-R1 ~88k stars","description":"DeepSeek (Hangzhou-based, Liang Wenfeng founder) — released DeepSeek-V3 671B MoE (37B active) Dec-2024, then R1 reasoning model Jan-2025 that matched o1 at ~3% of inference cost. Caused 'DeepSeek moment' in markets Jan-27-2025 — NVIDIA dropped 17%. MIT license. Substrate for cost-conscious OSS deployers and an active reasoning-research baseline.","cited_cells":57,"third_party_cells":9,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/deepseek-r1-v3-family--deepseek-com"},{"id":"deepspeed--gh-microsoft-deepspeed","name":"DeepSpeed","url":"https://github.com/deepspeedai/DeepSpeed","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-28","type":"ZeRO-based distributed training","license":"Apache-2.0","latest_release":"v0.19.8 (2026-10-20)","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"deepspeedai/DeepSpeed — ~43.3k stars, last commit 2026-10","description":"Microsoft's distributed-training library — ZeRO optimizer states/grads/params partitioning; ZeRO-Inference for serving. Foundation under many OSS LLM training stacks.","cited_cells":17,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/deepspeed--gh-microsoft-deepspeed"},{"id":"distilabel--gh-argilla-io-distilabel","name":"Distilabel","url":"https://github.com/argilla-io/distilabel","section":"Training infrastructure","tier":3,"status":"active","last_verified_at":"2026-10-27","type":"Synthetic-data pipelines for LLMs","license":"Apache-2.0","latest_release":"v1.5.3 (2025-01-28)","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"argilla-io/distilabel — ~3.4k stars, last commit 2026-10","description":"Argilla's framework for synthetic data generation — pipelines of LLM steps that produce SFT/DPO training data; integrates with Hub.","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/distilabel--gh-argilla-io-distilabel"},{"id":"dpo--arxiv-2305-18290","name":"DPO","url":"https://arxiv.org/abs/2305.18290","section":"Training infrastructure","tier":3,"status":"stale","last_verified_at":"2026-10-27","type":"Direct Preference Optimization","license":"Apache-2.0 (authors reference implementation)","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"eric-mitchell/direct-preference-optimization — ~2.9k stars, last commit 2024-08","description":"Stanford method that converts RLHF into a supervised-learning loss over preference pairs — no separate reward model needed; widely adopted alignment recipe.","cited_cells":10,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/dpo--arxiv-2305-18290"},{"id":"dvc--dvc-org","name":"DVC","url":"https://dvc.org/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Data version control","license":"Apache-2.0","latest_release":"v3.67.1 (2026-03-31)","deployment":"Self-hosted only — CLI and VS Code extension","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"treeverse/dvc — ~15.9k stars, last commit 2026-10","description":"Open-source data and ML model version control — git-like commands for datasets/models stored in cloud blob storage. Created by Iterative.ai; the project was acquired by lakeFS (Treeverse) in Nov 2025 and now lives at github.com/treeverse/dvc.","cited_cells":19,"third_party_cells":6,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/dvc--dvc-org"},{"id":"figure-ai--figure-ai","name":"Figure AI","url":"https://www.figure.ai/","section":"Robotics foundation models & agent stacks","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Humanoid robot maker (Figure 01 / Figure 02 / Helix VLA model)","license":"","latest_release":"Helix 2.5 (2026-09-17)","deployment":"Direct customer deployment (humanoid robot)","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"Internal Figure cloud","llm_lock":"","github":"","description":"Figure AI — humanoid robot maker founded 2022 by Brett Adcock (ex-Archer Aviation). Figure 02 (Aug 2024) + Helix VLA model (Feb 2025) for full-body / hand control. Raised $675M Series B Feb-2024 (Microsoft / OpenAI / Bezos / Intel; $2.6B val); reported new round at $39.5B val Q2 2025. Partnered with BMW for manufacturing pilots. Figure 03 (introduced 2025-10-09) is now marketed as a general-purpose home-help humanoid running Helix; Figure 02 was decommissioned (2026-09-30).","cited_cells":33,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/figure-ai--figure-ai"},{"id":"fireworks-ai--fireworks-ai","name":"Fireworks AI","url":"https://fireworks.ai/","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Fast OSS LLM inference cloud","license":"proprietary; Python SDK Apache-2.0","latest_release":"v1.2.30 (2026-10-21) — fireworks-ai","deployment":"searched not found","pricing":"Serverless pay-per-token; on-demand GPUs $8/hr (H100/H200), $13/hr (B200), $15/hr (B300); fine-tuning from $0.50 per 1M training tokens (LoRA SFT up to 16B)","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Fast inference platform for OSS LLMs — custom Cuda kernels, speculative decoding, multi-LoRA serving. Targets latency-sensitive production use cases.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/fireworks-ai--fireworks-ai"},{"id":"flyte--flyte-org","name":"Flyte","url":"https://flyte.org/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Data + ML workflow engine","license":"Apache-2.0","latest_release":"v2.0.53 (2026-10-27)","deployment":"Both (self-hosted + managed cloud) — OSS on your own Kubernetes; Union.ai hosted control plane","pricing":"Flyte OSS free; Union.ai Team $950/mo minimum credited to usage; Enterprise custom annual commit","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"flyteorg/flyte — ~7.8k stars, last commit 2026-10","description":"Kubernetes-native workflow engine — type-safe Python tasks; Lyft origin; Union AI commercial. Strong adoption in regulated industries.","cited_cells":18,"third_party_cells":6,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/flyte--flyte-org"},{"id":"fsdp-pytorch--pytorch-org","name":"FSDP (PyTorch)","url":"https://pytorch.org/docs/stable/fsdp.html","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Fully Sharded Data Parallel","license":"BSD-3-Clause (PyTorch)","latest_release":"v2.14.1 (2026-09-30) — torch","deployment":"Self-hosted only — library (part of PyTorch)","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"pytorch/pytorch — ~105k stars, last commit 2026-10","description":"Native PyTorch fully-sharded data parallel implementation — Meta's response to DeepSpeed ZeRO; FSDP2 in 2024 with improved API.","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/fsdp-pytorch--pytorch-org"},{"id":"galileo--galileo-ai","name":"Galileo","url":"https://www.galileo.ai/","section":"Evaluation & observability platforms","tier":1,"status":"acquired","last_verified_at":"2026-10-28","type":"Eval + observability + guardrails","license":"proprietary; Python SDK Apache-2.0","latest_release":"v2.7.0 (2026-10-14) — galileo","deployment":"Both (self-hosted + managed cloud) — SaaS, VPC or on-premises","pricing":"Splunk Agent Observability from $100/mo per 1.2M spans billed annually (includes 500M Luna tokens); free edition available","mcp_support":"yes — Agent Evals MCP server for IDEs such as Cursor and VS Code","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"rungalileo/galileo-python — ~22 stars, last commit 2026-10","description":"Galileo Evaluate / Observe / Protect — eval, tracing, guardrails; hallucination detection and ChainPoll metric. Acquired by Cisco in 2026 and now sold as Splunk Agent Observability.","cited_cells":21,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/galileo--galileo-ai"},{"id":"google-cloud-model-armor--cloud-google-com","name":"Google Cloud Model Armor","url":"https://cloud.google.com/security/products/model-armor","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Runtime security service that screens AI prompts, responses and agent interactions for prompt injection, jailbreaks, malicious URLs, sensitive-data leaks and harmful content, with adjustable confidence thresholds.","cited_cells":9,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/google-cloud-model-armor--cloud-google-com"},{"id":"google-deepmind-gemini-robotics--deepmind-google","name":"Google DeepMind Gemini Robotics","url":"https://deepmind.google/models/gemini-robotics/","section":"Robotics foundation models & agent stacks","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Robotics foundation model from Google DeepMind","license":"","latest_release":"Gemini Robotics 2 (with Gemini Robotics ER 2 and On-Device 2; date not stated on the model page)","deployment":"Trusted-tester via partners; Vertex AI / Cloud planned","pricing":"","mcp_support":"","a2a_support":"Google's own A2A protocol — likely first-party","backend_storage":"Google Cloud","llm_lock":"","github":"","description":"Google DeepMind's Gemini Robotics + Gemini Robotics-ER (Embodied Reasoning) released Mar-2025. VLA models built on Gemini 2.0 backbone. Reference platform: Apptronik Apollo. The DeepMind line of robotics-FM, descendant of earlier RT-1 / RT-2 / RT-X work.","cited_cells":59,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/google-deepmind-gemini-robotics--deepmind-google"},{"id":"google-gemini-3-family--blog-google","name":"Google Gemini 3 family","url":"https://blog.google/products-and-platforms/products/gemini/gemini-3/","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Frontier foundation model family (Gemini 3 Pro / Flash / Nano)","license":"","latest_release":"Gemini 4 Argon (GA 2026-10-13); Gemini 3.8 Flash (2026-09-02)","deployment":"Managed cloud (Google AI Studio + GCP Vertex AI); on-prem via Gemini-Distributed-Cloud + Gemma open-weights","pricing":"Paid per 1M in/out: Gemini 4 Argon $2/$10 intro to 2026-11-30, then $4/$20 (no free tier); Gemini 3.8 Flash $0.75/$3.75 (free tier available)","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"single-vendor (Google-only family)","github":"","description":"Google DeepMind's Gemini 3 frontier model family — Gemini 3 Pro (top tier), 3 Flash (fast), 3 Nano (on-device). Substrate for Google Workspace AI, Vertex AI Agent Platform, Gemini Code Assist, Project Mariner, and Project Astra. Multimodal-native; 1M-10M context window. Largest context window of any frontier model. Used by Gemini Enterprise Agent Platform Memory Bank (separate row).","cited_cells":34,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/google-gemini-3-family--blog-google"},{"id":"google-vertex-ai-training--cloud-google-com","name":"Google Vertex AI Training (now Gemini Enterprise Agent Platform training)","url":"https://cloud.google.com/products/gemini-enterprise-agent-platform","section":"Training infrastructure","tier":1,"status":"renamed","last_verified_at":"2026-10-29","type":"GCP ML training","license":"proprietary; Python SDK Apache-2.0","latest_release":"v2.6.0 (2026-10-20) — google-cloud-aiplatform","deployment":"Managed-only — Google Cloud service","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"googleapis/python-aiplatform — ~910 stars, last commit 2026-10","description":"Google Cloud's managed training service — TPU and GPU clusters, hyperparameter tuning, model registry. Part of Vertex AI umbrella. Vertex AI is now branded Gemini Enterprise Agent Platform (Gemini platform), and training is documented there as serverless and custom training (checked 2026-10-29).","cited_cells":15,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/google-vertex-ai-training--cloud-google-com"},{"id":"grpo--arxiv-2402-03300","name":"GRPO","url":"https://arxiv.org/abs/2402.03300","section":"Training infrastructure","tier":3,"status":"active","last_verified_at":"2026-10-27","type":"Group Relative Policy Optimization","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","description":"DeepSeek-Math's GRPO — group-relative advantage estimation replacing PPO's critic. Used in DeepSeek-R1 reasoning post-training; widely adopted in OSS reasoning RLHF.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/grpo--arxiv-2402-03300"},{"id":"guardrails-ai--guardrailsai-com","name":"Guardrails AI","url":"https://guardrailsai.com/","section":"Guardrails & AI security","tier":3,"status":"acquired","last_verified_at":"2026-10-29","type":"Framework","license":"Apache-2.0","latest_release":"last commit 2026-08-26","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/guardrails-ai/guardrails (7568★)","description":"Open-source guardrails framework plus the Guardrails Hub of validators that detect policy violations, hallucinations and data leakage in LLM and agent outputs at runtime. Harvey announced its acquisition of Guardrails AI on Sep 8, 2026.","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/guardrails-ai--guardrailsai-com"},{"id":"helicone--helicone-ai","name":"Helicone","url":"https://www.helicone.ai/","section":"Evaluation & observability platforms","tier":2,"status":"maintenance","last_verified_at":"2026-10-29","type":"LLM observability + gateway","license":"Apache-2.0 (Core)","latest_release":"v1.8.3 (2025-11-07) — @helicone/helpers","deployment":"Both (self-hosted + managed cloud) — Docker Compose self-host; Helm and on-prem for Enterprise","pricing":"Hobby free (10k requests); Pro $79/mo; Team $799/mo; Enterprise custom; plus usage-based pricing","mcp_support":"yes — @helicone/mcp server to query requests and sessions from MCP clients","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"Helicone/helicone — ~6.2k stars, last commit 2026-10","description":"Open-source LLM observability and AI gateway — one-line proxy integration captures every LLM call; sessions, custom properties, evals. Acquired by Mintlify in Mar 2026 and placed in maintenance mode; helicone.ai now states it is no longer part of Mintlify.","cited_cells":22,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/helicone--helicone-ai"},{"id":"honeyhive--honeyhive-ai","name":"HoneyHive","url":"https://www.honeyhive.ai/","section":"Evaluation & observability platforms","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Agent observability and evaluation platform for replaying traces step by step, debugging long-running multi-agent trajectories and evaluating live production traffic.","cited_cells":14,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/honeyhive--honeyhive-ai"},{"id":"hugging-face-datasets--huggingface-co","name":"Hugging Face Datasets","url":"https://huggingface.co/docs/datasets","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Dataset library + Hub","license":"Apache-2.0","latest_release":"v5.1.0 (2026-10-05)","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"huggingface/datasets — ~22.1k stars, last commit 2026-10","description":"Library for accessing and processing 100k+ public datasets on the Hub. Apache Arrow-backed; default dataset layer for the open ML ecosystem.","cited_cells":16,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/hugging-face-datasets--huggingface-co"},{"id":"hugging-face-lerobot--huggingface-co","name":"Hugging Face LeRobot","url":"https://huggingface.co/lerobot","section":"Robotics foundation models & agent stacks","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"OSS robotics platform on HuggingFace (datasets + models + policies)","license":"Apache-2.0","latest_release":"v0.6.1 (2026-08-03)","deployment":"Self-hosted only — library (pip install lerobot)","pricing":"OSS free; HF Hub free + paid tiers","mcp_support":"","a2a_support":"","backend_storage":"HF Hub for datasets / model weights","llm_lock":"","github":"huggingface/lerobot — ~28.5k stars, last commit 2026-10","description":"Hugging Face's open-source robotics platform — datasets, baselines, training code for imitation learning + diffusion policy + ACT. Launched May 2024; growing rapidly with community contributions (HopeJR, SO-100 arms). LeRobot is the de-facto OSS robotics-ML platform analogous to transformers.","cited_cells":71,"third_party_cells":5,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/hugging-face-lerobot--huggingface-co"},{"id":"humanloop--humanloop-com","name":"Humanloop","url":"https://humanloop.com/","section":"Evaluation & observability platforms","tier":2,"status":"archived","last_verified_at":"2026-10-29","type":"LLM eval + collab","license":"searched not found","latest_release":"v0.8.38 (2025-05-09) — humanloop SDK, final","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"humanloop/humanloop-python — ~12 stars, last commit 2025-05","description":"Collaborative LLM development platform — prompt editor, evals, tracing. UK; Index Ventures + Y Combinator backed. The team joined Anthropic in 2025 and the platform shut down on 2025-09-08.","cited_cells":13,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/humanloop--humanloop-com"},{"id":"inspect-ai--inspect-aisi-org-uk","name":"Inspect AI","url":"https://inspect.aisi.org.uk/","section":"Evaluation & observability platforms","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"MIT","latest_release":"last commit 2026-10-29","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/UKGovernmentBEIS/inspect_ai (3128★)","description":"Open-source framework for frontier LLM evaluations from the UK AI Security Institute and Meridian Labs. It has composable datasets, agents, tools and scorers, 200+ prebuilt evals including agentic tasks, and a log viewer.","cited_cells":14,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/inspect-ai--inspect-aisi-org-uk"},{"id":"invariant-guardrails--invariantlabs-ai","name":"Invariant Guardrails","url":"https://invariantlabs.ai/blog/guardrails","section":"Guardrails & AI security","tier":3,"status":"acquired","last_verified_at":"2026-10-29","type":"Framework","license":"Apache-2.0","latest_release":"last commit 2026-01-12","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/invariantlabs-ai/invariant (475★)","description":"Contextual guardrailing layer for agentic and MCP applications that sits at the LLM and MCP level. It enforces deterministic rules over tool calls, data flow, secrets, PII, code and loops.","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/invariant-guardrails--invariantlabs-ai"},{"id":"jamba-ai21-labs--ai21-com","name":"Jamba (AI21 Labs)","url":"https://www.ai21.com/jamba/","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-22","type":"SSM-Transformer hybrid + 256K KV-bypass context","license":"Apache-2.0 (Jamba2); earlier Jamba 1.5–1.7 under Jamba Open Model License","latest_release":"Jamba2 3B / Jamba2 Mini (52B, 12B active) (2026-01-08)","deployment":"Both (AI21 Studio cloud; Azure/AWS Bedrock/GCP Vertex AI; private VPC; on-premises)","pricing":"Free trial ($10 credit, no CC required); pay-as-you-go per token; enterprise custom pricing with VPC/on-prem","mcp_support":"not documented publicly","a2a_support":"not documented publicly","backend_storage":"searched not found","llm_lock":"","github":"not applicable — open weights on Hugging Face (huggingface.co/ai21labs)","description":"Israeli production model. Interleaves Mamba SSM layers with Transformer attention (1:7 ratio) + MoE blocks every 2 layers. SSM layers maintain fixed-size recurrent hidden state — compressed long-context memory bypassing growing KV cache.","cited_cells":40,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/jamba-ai21-labs--ai21-com"},{"id":"k-scale-labs--kscale-dev","name":"K-Scale Labs","url":"https://www.kscale.dev/","section":"Robotics foundation models & agent stacks","tier":3,"status":"stale","last_verified_at":"2026-10-27","type":"Open-source humanoid + OS — Y Combinator W24","license":"MIT (kos); K-Bot software GPL-3.0, hardware CERN-OHL-S","latest_release":"v0.3.18 (2025-09-04) — kscale SDK","deployment":"Hardware ships; OS self-host","pricing":"Hardware kits ($8-16k range); OS free","mcp_support":"","a2a_support":"","backend_storage":"Caller-managed","llm_lock":"","github":"kscalelabs/kos — ~75 stars, last commit 2025-05","description":"Open-source humanoid robot maker — K-Scale OS (Kos), Stompy and KBot humanoids. YC W24. Pitched as the 'open-source humanoid' alternative to Figure / Tesla — full hardware + software OSS.","cited_cells":37,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/k-scale-labs--kscale-dev"},{"id":"label-studio-humansignal--labelstud-io","name":"Label Studio (HumanSignal)","url":"https://labelstud.io/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Data labelling platform","license":"Apache-2.0 (Community); proprietary Enterprise","latest_release":"v1.23.2 (2026-09-29)","deployment":"Both (self-hosted + managed cloud) — Community self-host; Starter Cloud hosted; Enterprise cloud or on-prem","pricing":"Community free; Starter Cloud $99/mo for up to 12 users (+$49/user); Enterprise custom","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"HumanSignal/label-studio — ~28.6k stars, last commit 2026-10","description":"Open-source data-labelling platform supporting text, image, audio, video, time-series. Commercial Enterprise edition + Cloud (HumanSignal).","cited_cells":19,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/label-studio-humansignal--labelstud-io"},{"id":"labelbox--labelbox-com","name":"Labelbox","url":"https://labelbox.com/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Data factory for AI","license":"proprietary; Python SDK Apache-2.0","latest_release":"v7.17.0 (2026-10-22) — labelbox SDK","deployment":"searched not found","pricing":"searched not found","mcp_support":"yes — documented MCP client connection for project insights and approved actions","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"Labelbox/labelbox-python — ~158 stars, last commit 2026-10","description":"Originally an enterprise data-labelling platform (annotation tools plus expert RLHF data for frontier labs). By 2026-10 it sells Horizon reinforcement-learning environments (WorldSim) to frontier AI labs and Recursion managed agents to enterprises.","cited_cells":18,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/labelbox--labelbox-com"},{"id":"lakefs--lakefs-io","name":"lakeFS","url":"https://lakefs.io/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Git-like data lake","license":"BSL-1.1 from v1.87.0 (2026-09-22; each release converts to Apache-2.0 after 4 years); earlier releases Apache-2.0","latest_release":"v1.88.0 (2026-10-01)","deployment":"Both (self-hosted + managed cloud) — self-managed; managed lakeFS Cloud with Enterprise","pricing":"lakeFS Team $499/mo (5 users, 500GB); Enterprise custom","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"treeverse/lakeFS — ~5.6k stars, last commit 2026-10","description":"Git-like branching, commits and rollback for object-store data lakes (S3/GCS/Azure). Built by Treeverse, which also acquired DVC in Nov 2025. The Community edition is source-available under BSL 1.1 from v1.87.0 (Sep 2026); Enterprise and lakeFS Cloud are commercial.","cited_cells":18,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/lakefs--lakefs-io"},{"id":"laminar--laminar-sh","name":"Laminar","url":"https://laminar.sh/","section":"Evaluation & observability platforms","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"Apache-2.0","latest_release":"last commit 2026-10-16","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/lmnr-ai/lmnr (3460★)","description":"Open-source agent observability platform that captures LLM calls, tool calls, sub-agents, costs and tokens. Its Signals feature clusters agent failures automatically, and it also runs evals.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/laminar--laminar-sh"},{"id":"litellm--gh-berriai-litellm","name":"LiteLLM","url":"https://github.com/BerriAI/litellm","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Universal LLM API + Proxy","license":"MIT (except enterprise/ directory, commercial license)","latest_release":"v1.106.4 (2026-10-25)","deployment":"Self-hosted only — SDK and proxy; Enterprise self-hosted or air-gapped","pricing":"OSS free; Enterprise custom pricing (30-day self-hosted trial key)","mcp_support":"yes — MCP Gateway in LiteLLM Proxy (Streamable HTTP, SSE, stdio) with per-key/team access control","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"BerriAI/litellm — ~62.1k stars, last commit 2026-10","description":"BerriAI's open-source LLM gateway — unified OpenAI-format API for 100+ providers; LiteLLM Proxy adds budgets, fallbacks, observability, rate-limiting.","cited_cells":26,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/litellm--gh-berriai-litellm"},{"id":"llama-factory--gh-hiyouga-llama-factory","name":"LLaMA-Factory","url":"https://github.com/hiyouga/LlamaFactory","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Unified LLM fine-tuning toolbox","license":"Apache-2.0","latest_release":"v0.9.5 (2026-05-30)","deployment":"Self-hosted only — library with CLI and WebUI","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"hiyouga/LlamaFactory — ~75.7k stars, last commit 2026-10","description":"Unified efficient fine-tuning of 100+ LLMs — SFT, DPO, PPO, KTO, ORPO with LoRA/QLoRA/full; CLI + WebUI. Popular in the Chinese OSS scene.","cited_cells":16,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/llama-factory--gh-hiyouga-llama-factory"},{"id":"lora--arxiv-2106-09685","name":"LoRA","url":"https://arxiv.org/abs/2106.09685","section":"Training infrastructure","tier":3,"status":"stale","last_verified_at":"2026-10-27","type":"Low-rank adapter fine-tuning","license":"MIT (microsoft/LoRA, loralib)","latest_release":"v0.1.2 (2023-08-27) — loralib","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"microsoft/LoRA — ~13.9k stars, last commit 2024-12","description":"Microsoft Research's Low-Rank Adapters method — trains rank-r matrices added to attention weights; 10000x parameter reduction. Foundation of all adapter-style fine-tuning.","cited_cells":13,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/lora--arxiv-2106-09685"},{"id":"megatron-core--gh-nvidia-megatron-lm","name":"Megatron-Core","url":"https://github.com/NVIDIA/Megatron-LM","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Library-form of Megatron","license":"BSD-3-Clause","latest_release":"v0.19.2 (2026-09-18) — megatron-core","deployment":"Self-hosted only — library (pip install megatron-core)","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"NVIDIA/Megatron-LM — ~18.3k stars, last commit 2026-10","description":"PyTorch-library form of Megatron-LM — distributed training building blocks (tensor/pipeline/expert parallel) packaged for embedding into custom training stacks.","cited_cells":15,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/megatron-core--gh-nvidia-megatron-lm"},{"id":"megatron-lm--gh-nvidia-megatron-lm","name":"Megatron-LM","url":"https://github.com/NVIDIA/Megatron-LM","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Tensor/pipeline parallel LLM training","license":"BSD-3-Clause","latest_release":"v0.19.2 (2026-09-18) — megatron-core","deployment":"Self-hosted only — reference training scripts plus library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"NVIDIA/Megatron-LM — ~18.3k stars, last commit 2026-10","description":"NVIDIA's reference framework for training large transformer models with tensor, pipeline, sequence, and expert parallelism. Underpins many frontier-lab training stacks.","cited_cells":16,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/megatron-lm--gh-nvidia-megatron-lm"},{"id":"meta-llama-4-family--ai-meta-com","name":"Meta Llama 4 family","url":"https://ai.meta.com/blog/llama-4-multimodal-intelligence/","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Open-weights frontier model family (Llama 4 Scout / Maverick / Behemoth)","license":"Llama 4 community license (Meta-custom; commercial OK <700M MAU)","latest_release":"No Llama release found after Llama 4 Scout + Maverick (2025-04-05); successor Muse family: Muse Spark (2026-04-08, closed), Muse Spark Open weights (2026-10-15)","deployment":"Both (self-hosted + managed cloud) — open weights (llama.com / Hugging Face); hosted via AWS Bedrock, Azure AI, Together, Groq, Fireworks, Databricks","pricing":"Free for self-hosting (Llama community license, <700M MAU); hosted via Together $0.20-$3/M (Maverick); AWS Bedrock + Azure AI consumption","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"open (Llama community license; many hosted providers)","github":"github.com/meta-llama/llama (combined Llama repos); ~58k stars","description":"Meta's Llama 4 family (released 2025-04) — Scout (109B MoE, 17B active), Maverick (400B MoE, 17B active), Behemoth (2T MoE, 288B active; preview). First Llama generation to use mixture-of-experts and to be multimodal-native. Substrate for many OSS agents (Open-Interpreter, Ollama-served local agents, vLLM-served enterprise inference). Llama community license (commercial use permitted under 700M MAU threshold).","cited_cells":35,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/meta-llama-4-family--ai-meta-com"},{"id":"meta-llama-protections-llama-guard-prompt-guard-llamafirewall--dev-meta-ai","name":"Meta Llama Protections (Llama Guard, Prompt Guard, LlamaFirewall)","url":"https://dev.meta.ai/llama/llama-protections","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"Llama 3.2 Community License (Llama Guard / Prompt Guard models); LlamaFirewall and Code Shield MIT","latest_release":"last commit 2026-10-12","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/meta-llama/PurpleLlama (4450★)","description":"Meta's open system-level safeguards from the Purple Llama project: Llama Guard content-moderation classifiers, Prompt Guard prompt-injection and jailbreak detectors, Code Shield insecure-code filtering, and LlamaFirewall, a scanner framework for securing multi-step agentic operations.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/meta-llama-protections-llama-guard-prompt-guard-llamafirewall--dev-meta-ai"},{"id":"microsoft-phi-4-family--azure-microsoft-com","name":"Microsoft Phi-4 family","url":"https://azure.microsoft.com/en-us/products/phi","section":"Foundation models (substrate reference)","tier":2,"status":"active","last_verified_at":"2026-10-31","type":"Small / efficient open-weights model family (Phi-4 14B + Phi-4 multimodal + Phi-4 mini)","license":"MIT","latest_release":"Phi-4-reasoning-vision-15B (2026-03)","deployment":"Both (self-hosted + managed cloud) — MIT open weights (Hugging Face, vLLM/Transformers); Microsoft Foundry catalog; on-device via ONNX Runtime","pricing":"Free open-weights (MIT); Azure AI pay-per-token if hosted","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"open (MIT)","github":"github.com/microsoft/Phi-3CookBook ~3k+ stars + various HuggingFace repos","description":"Microsoft Research's Phi family of small, data-curation-focused open-weights models. Phi-4 (14B, Dec-2024) matches much larger models on STEM benchmarks. Phi-4 Multimodal (5.6B, Feb-2025) integrates speech + vision + text. MIT license. Substrate for on-device / edge agents and cost-sensitive OSS use.","cited_cells":40,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/microsoft-phi-4-family--azure-microsoft-com"},{"id":"mistral-large-2-mixtral-family--mistral-ai","name":"Mistral Large 2 / Mixtral family","url":"https://mistral.ai/news/mistral-large-2407/","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Frontier model family — Mistral Large 2 + Mixtral 8x22B + Mistral Small 3","license":"Mixed: Mistral Research License MRL-0.1, non-OSI (Mistral Large 2 / 2407 and 2411 weights; commercial use needs a separate licence); Mistral Non-Production License (Codestral 22B); Modified MIT (Medium 3.5 weights); Apache-2.0 (Mixtral 8x22B, Small / Large 3 / other open models); some models API-only proprietary","latest_release":"Mistral Medium 3.5 (2026-04-29); Mistral Large 4 in public preview (2026-10), weights not yet released","deployment":"Both (self-hosted + managed cloud) — La Plateforme / Mistral Studio API, open weights for self-hosting, plus Azure / AWS / GCP marketplaces","pricing":"API: Mistral Medium 3.5 $1.50 in / $7.50 out per 1M; Mistral Large 3 $0.50 / $1.50; batch 50% off; open weights free","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"open (multi-vendor — Apache 2.0 weights + hosted)","github":"github.com/mistralai/mistral-inference + various model repos; cumulative ~30k+ stars","description":"French frontier-model lab (Paris). Family includes Mistral Large 2 (123B dense, Jul-2024), Mixtral 8x22B (open-weights MoE), Mistral Small 3 (24B, open-weights Jan-2025), Codestral (code specialist). EU-headquartered alternative to US/China model labs; GDPR-friendly data residency. Distinct from the Mistral Le Chat memory feature (already in catalog) — this row covers the model family itself as substrate.","cited_cells":37,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mistral-large-2-mixtral-family--mistral-ai"},{"id":"mlflow--mlflow-org","name":"MLflow","url":"https://mlflow.org/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Open ML lifecycle platform","license":"Apache-2.0","latest_release":"v3.17.2 (2026-10-21) — mlflow","deployment":"searched not found","pricing":"searched not found","mcp_support":"yes — built-in MCP server (experimental, mlflow mcp run, MLflow 3.5.1+) exposing trace search, tagging and feedback tools","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"mlflow/mlflow — ~28.6k stars, last commit 2026-10","description":"Databricks-incubated open platform for ML lifecycle — tracking, projects, models, model registry. Standard ML-ops baseline. MLflow 3.0 (2024) added LLM-eval features.","cited_cells":17,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mlflow--mlflow-org"},{"id":"modal--modal-com","name":"Modal","url":"https://modal.com/","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Serverless compute / inference","license":"proprietary; client SDK Apache-2.0","latest_release":"v1.6.2 (2026-10-16) — modal","deployment":"Managed-only — serverless cloud","pricing":"Starter $0 + compute ($30/mo free credits); Team $250/mo + compute ($100/mo credits); Enterprise custom; usage per second (e.g. H100 $0.001097/sec)","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"modal-labs/modal-client — ~530 stars, last commit 2026-10","description":"Serverless cloud for AI/ML — Python-decorator workflow defines GPU/CPU functions deployed to managed infra; widely used for training jobs, batch inference, and agent backends.","cited_cells":22,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/modal--modal-com"},{"id":"nemotron-synthetic-data-nvidia--blogs-nvidia-com","name":"Nemotron Synthetic Data (NVIDIA)","url":"https://blogs.nvidia.com/blog/nemotron-4-synthetic-data-generation-llm-training/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Synthetic SFT/RLHF data pipelines","license":"NVIDIA Open Model License — Nemotron-4 340B weights","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"searched not found","description":"NVIDIA's open synthetic-data generation pipelines (HelpSteer2, NemoSkills, Nemotron-4 340B). Used by frontier labs to bootstrap RLHF reward modeling data.","cited_cells":8,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/nemotron-synthetic-data-nvidia--blogs-nvidia-com"},{"id":"nvidia-gr00t-isaac--developer-nvidia-com","name":"NVIDIA GR00T / Isaac","url":"https://www.nvidia.com/en-us/industries/robotics/isaac-groot/","section":"Robotics foundation models & agent stacks","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Robotics foundation model + simulation stack (NVIDIA)","license":"Apache-2.0 code; model weights NVIDIA Open Model License","latest_release":"GR00T N1.7 (2026-04-21) — Isaac-GR00T","deployment":"On-robot (Jetson Thor) + simulation (Isaac Sim) + training cloud (Cosmos)","pricing":"OSS weights free; Cosmos / Jetson hardware sold separately","mcp_support":"","a2a_support":"","backend_storage":"Caller-managed (on-robot + cloud)","llm_lock":"","github":"NVIDIA/Isaac-GR00T — ~8.3k stars, last commit 2026-10","description":"NVIDIA Project GR00T (Mar-2024) — general-purpose humanoid foundation model + Isaac simulation stack + Jetson Thor compute. GR00T N1 model released open March 2025. NVIDIA's full-stack play: model + sim + compute + training cloud (Cosmos). Major partners include Figure, Apptronik, Boston Dynamics, Agility.","cited_cells":71,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/nvidia-gr00t-isaac--developer-nvidia-com"},{"id":"nvidia-nemo-guardrails--docs-nvidia-com","name":"NVIDIA NeMo Guardrails","url":"https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"Apache-2.0","latest_release":"last commit 2026-10-29","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/NVIDIA-NeMo/Guardrails (7394★)","description":"Open-source Python library for adding programmable guardrails to LLM-based applications and agents. It blocks, alters or validates unsafe, off-topic, malicious or policy-violating user inputs and model responses using configuration files, Colang flows and built-in rails.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/nvidia-nemo-guardrails--docs-nvidia-com"},{"id":"octoai-now-nvidia--octo-ai","name":"OctoAI (now NVIDIA)","url":"https://octo.ai/","section":"Inference platforms & gateways","tier":2,"status":"acquired","last_verified_at":"2026-10-29","type":"GenAI inference (acquired)","license":"searched not found","latest_release":"octoai Python SDK 1.11.0 (2024-09-18) — final release before shutdown","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Inference cloud spun out of OctoML — acquired by NVIDIA in 2024. Originally TVM-compiler heritage; offered API for OSS LLM/diffusion serving.","cited_cells":9,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/octoai-now-nvidia--octo-ai"},{"id":"openai-evals--gh-openai-evals","name":"OpenAI Evals","url":"https://github.com/openai/evals","section":"Evaluation & observability platforms","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"MIT","latest_release":"last commit 2026-04-14","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/openai/evals (19675★)","description":"OpenAI's open-source framework and registry of benchmarks for evaluating LLMs and LLM systems.","cited_cells":13,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openai-evals--gh-openai-evals"},{"id":"openai-gpt-family-gpt-5-gpt-4o-o3-o4--openai-com","name":"OpenAI GPT family (GPT-5 / GPT-4o / o3 / o4)","url":"https://openai.com/index/introducing-gpt-5/","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Frontier foundation model family (GPT-5 / GPT-4o / o3 / o4)","license":"proprietary (API only)","latest_release":"GPT-6.1 Terra (2026-10-26; computer use) and GPT-6.1 Sol; earlier GPT-6 Sol / Luna / Instant in ChatGPT (2026-10-07) and API flagship GPT-6 Astra (announced 2026-09-03)","deployment":"Managed-only — OpenAI API + ChatGPT, also Microsoft Azure and AWS Bedrock (separate gpt-oss open-weight models exist)","pricing":"API flagship GPT-6 Astra $10 input / $50 output per 1M tokens (Fast mode 2x price); ChatGPT Free / Go / Plus / Pro / Business / Enterprise plans","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"single-vendor (OpenAI-only family)","github":"","description":"OpenAI's foundation model family — GPT-5 (2025-08 flagship), GPT-4o (multimodal workhorse), o3/o4 reasoning series. Substrate for ChatGPT, the OpenAI API, Microsoft Copilot, Azure OpenAI, and many downstream agents in this catalog (Devin, Codex CLI, Operator, Cursor mode). Parametric memory in weights; discrete user memory ships in ChatGPT Memory feature (separate row).","cited_cells":36,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openai-gpt-family-gpt-5-gpt-4o-o3-o4--openai-com"},{"id":"openinference--gh-arize-ai-openinference","name":"OpenInference","url":"https://github.com/Arize-ai/openinference","section":"Evaluation & observability platforms","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"OTel semantic conventions for LLMs","license":"Apache-2.0","latest_release":"v0.1.42 (2026-10-23) — openinference-semantic-conventions","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"Arize-ai/openinference — ~1.3k stars, last commit 2026-10","description":"Arize-led open OTel semantic conventions for LLM/agent tracing — auto-instrumentation for OpenAI/Anthropic/LangChain/LlamaIndex etc.","cited_cells":19,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openinference--gh-arize-ai-openinference"},{"id":"openllmetry-traceloop--gh-traceloop-openllmetry","name":"OpenLLMetry (Traceloop)","url":"https://github.com/traceloop/openllmetry","section":"Evaluation & observability platforms","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"OTel auto-instrumentation for LLMs","license":"Apache-2.0","latest_release":"v0.62.4 (2026-09-29) — traceloop-sdk","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"traceloop/openllmetry — ~7.5k stars, last commit 2026-10","description":"Traceloop's open-source OpenTelemetry auto-instrumentation for LLM workflows — works with any OTel-compatible backend (Datadog, New Relic, Grafana, etc.).","cited_cells":21,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openllmetry-traceloop--gh-traceloop-openllmetry"},{"id":"openmind-om1--openmind-org","name":"OpenMind OM1","url":"https://openmind.com/","section":"Robotics foundation models & agent stacks","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Open robot OS / agentic framework (FABRIC + OM1)","license":"MIT","latest_release":"OM1 (open robot operating model)","deployment":"Self-hosted only — prebuilt binaries for Linux/macOS or build from source; connects to robot HAL/ROS2 via DDS or websockets","pricing":"OSS free; commercial support paid","mcp_support":"","a2a_support":"","backend_storage":"Caller-managed","llm_lock":"","github":"OpenMind/OM1 — ~3.0k stars, last commit 2026-10","description":"OpenMind — open robot OS (OM1) + FABRIC interconnect, founded by Jan Liphardt (Stanford). Pitched as the 'Android of robots': any humanoid + any LLM + the same OS layer. Raised $20M Series A (Pantera + others, 2025).","cited_cells":42,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openmind-om1--openmind-org"},{"id":"openrlhf--gh-openrlhf-openrlhf","name":"OpenRLHF","url":"https://github.com/OpenRLHF/OpenRLHF","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Distributed RLHF training framework","license":"Apache-2.0","latest_release":"v0.11.2 (2026-09-14) — openrlhf","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"OpenRLHF/OpenRLHF — ~10.1k stars, last commit 2026-10","description":"Distributed RLHF training framework built on Ray + DeepSpeed + vLLM — supports PPO/DPO/REINFORCE++/GRPO. Targeted at 70B+ models with efficient sample throughput.","cited_cells":16,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openrlhf--gh-openrlhf-openrlhf"},{"id":"openrouter--openrouter-ai","name":"OpenRouter","url":"https://openrouter.ai/","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-26","type":"Multi-provider LLM gateway","license":"searched not found","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Unified API + marketplace for 300+ models — pay-as-you-go credits, automatic fallback, model arena. Used as default LLM router by many JS/TS agents.","cited_cells":13,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openrouter--openrouter-ai"},{"id":"outerbounds--outerbounds-com","name":"Outerbounds","url":"https://outerbounds.com/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-29","type":"Metaflow-based ML platform","license":"proprietary; Metaflow Apache-2.0","latest_release":"v0.12.53 (2026-10-16) — outerbounds","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"Netflix/metaflow — ~10.3k stars, last commit 2026-10","description":"Commercial platform behind Metaflow (originally Netflix OSS) — ML workflow orchestration with managed compute. Acquisition by Anaconda announced 2026-04-29; now marketed as AI Orchestration in Anaconda Platform, and Metaflow stays open source.","cited_cells":13,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/outerbounds--outerbounds-com"},{"id":"pachyderm-hpe--pachyderm-com","name":"Pachyderm (HPE)","url":"https://www.pachyderm.com/","section":"Training infrastructure","tier":2,"status":"acquired","last_verified_at":"2026-10-29","type":"Data versioning + pipelines","license":"Apache-2.0","latest_release":"v2.12.2 (2025-01-15)","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"pachyderm/pachyderm — ~6.3k stars, last commit 2025-02","description":"Data-versioning and pipeline platform — Pachyderm File System tracks lineage across pipeline runs. Acquired by HPE in 2023.","cited_cells":17,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/pachyderm-hpe--pachyderm-com"},{"id":"patronus-ai--patronus-ai","name":"Patronus AI","url":"https://www.patronus.ai/","section":"Evaluation & observability platforms","tier":2,"status":"archived","last_verified_at":"2026-10-29","type":"Enterprise eval + guardrails","license":"proprietary; Python SDK MIT","latest_release":"v0.1.26 (2026-09-17) — patronus","deployment":"searched not found","pricing":"Contact sales — no public price list","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Enterprise eval and guardrails platform — Lynx hallucination detector (open weights), Glider judge model, Polyglot benchmark. Built by ex-Meta researchers. The evaluation platform (API, SDKs, evaluators, Lynx, GLIDER, Percival) has been discontinued per its docs (checked 2026-10-29); Patronus AI now operates as a behavioral research lab building training data and simulated environments for agents.","cited_cells":13,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/patronus-ai--patronus-ai"},{"id":"peft-hugging-face--gh-huggingface-peft","name":"PEFT (Hugging Face)","url":"https://github.com/huggingface/peft","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Parameter-efficient fine-tuning","license":"Apache-2.0","latest_release":"v0.21.2 (2026-10-01) — peft","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"huggingface/peft — ~21.9k stars, last commit 2026-10","description":"Hugging Face's PEFT library — LoRA, AdaLoRA, IA3, Prefix Tuning, P-Tuning. Standard interface for adapter-style fine-tuning.","cited_cells":18,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/peft-hugging-face--gh-huggingface-peft"},{"id":"physical-intelligence--physicalintelligence-company","name":"Physical Intelligence (π)","url":"https://www.pi.website/","section":"Robotics foundation models & agent stacks","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Robotics foundation-model lab (π0 / π0.5)","license":"OpenPI Apache 2.0 (Feb 2025 release of weights)","latest_release":"pi0.7 (2026-04-16) — model; openpi repo has no GitHub releases","deployment":"Both (self-hosted + managed cloud) — openpi weights self-hosted; Pi API beta (managed fine-tuning + hosted inference, waitlist) since 2026-10-13","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"Caller manages robot trajectories","llm_lock":"","github":"Physical-Intelligence/openpi — ~14.4k stars, last commit 2026-08","description":"Founded 2024 by Sergey Levine (UC Berkeley) and Karol Hausman + others (ex-Google Brain Robotics). Building general-purpose robot foundation models (π0 released Oct-2024; π0.5 in 2025) trained on cross-embodiment robot trajectories. Raised $400M Series A (Nov-2024; Jeff Bezos, OpenAI, Thrive, Lux) at $2.4B val. Released π0.7 (2026-04-16) and opened its models through the Pi API beta (2026-10-13).","cited_cells":62,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/physical-intelligence--physicalintelligence-company"},{"id":"pickle-robot--picklerobot-com","name":"Pickle Robot","url":"https://www.picklerobot.com/","section":"Robotics foundation models & agent stacks","tier":3,"status":"active","last_verified_at":"2026-10-27","type":"Warehouse-truck-unloading robot platform (MIT spinout)","license":"","latest_release":"","deployment":"Truck-dock installation; RaaS","pricing":"RaaS / capex","mcp_support":"","a2a_support":"","backend_storage":"Internal cloud","llm_lock":"","github":"","description":"MIT spinout focused on truck-unloading robots — autonomously empties truck trailers. Raised $50M Series B (Mar 2024, Tiger Global + others). Narrow focus but production deployments at major logistics customers.","cited_cells":35,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/pickle-robot--picklerobot-com"},{"id":"portkey-ai--portkey-ai","name":"Portkey AI","url":"https://portkey.ai/","section":"Inference platforms & gateways","tier":2,"status":"acquired","last_verified_at":"2026-10-29","type":"Production AI gateway","license":"MIT (Gateway)","latest_release":"v1.15.2 (2026-01-12) — gateway","deployment":"Both (self-hosted + managed cloud) — OSS gateway self-hosted or Portkey cloud","pricing":"Open-source gateway free (self-host); Developer free (10K logs/mo); Production $49/mo (100K logs, $9 per extra 100K up to 3M); Enterprise custom","mcp_support":"yes — MCP Gateway product (unified layer to access and govern MCP tools)","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"Portkey-AI/gateway — ~13.3k stars, last commit 2026-05","description":"AI gateway + observability — load balancing across 200+ LLMs, retries, caching, eval, prompts. Open-source Gateway + commercial Portkey Cloud. Acquired by Palo Alto Networks (closed about 2026-06); now sold as Prisma AIRS AI Gateway (GA 2026-07-16).","cited_cells":23,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/portkey-ai--portkey-ai"},{"id":"prefect--prefect-io","name":"Prefect","url":"https://www.prefect.io/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Workflow orchestration","license":"Apache-2.0","latest_release":"v3.9.0 (2026-10-21) — prefect","deployment":"Both (self-hosted + managed cloud) — OSS server or Prefect Cloud","pricing":"Prefect Cloud: Hobby free; Starter $100/mo; Team $100/user/mo; Enterprise custom","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"PrefectHQ/prefect — ~24.1k stars, last commit 2026-10","description":"Python-native workflow orchestrator — widely used for ML pipelines and ETL. Prefect Cloud + open-source Prefect Core.","cited_cells":20,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/prefect--prefect-io"},{"id":"presidio--presidio-dataprivacystack-org","name":"Presidio","url":"https://presidio.dataprivacystack.org/","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"MIT","latest_release":"last commit 2026-10-29","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/data-privacy-stack/presidio (11579★)","description":"Open-source framework for detecting, redacting, masking and anonymizing PII in text, images and structured data, commonly used to scrub prompts and responses around LLMs. Originally a Microsoft project, it has transitioned to community ownership under Data Privacy Stack.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/presidio--presidio-dataprivacystack-org"},{"id":"prompt-security-sentinelone--sentinelone-com","name":"Prompt Security (SentinelOne)","url":"https://www.sentinelone.com/platform/securing-ai-prompt/","section":"Guardrails & AI security","tier":3,"status":"acquired","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"AI security platform that controls employee AI tool usage, protects homegrown AI apps and autonomous agents, and redacts sensitive data and enforces policy across AI services. It is now offered by SentinelOne, and prompt.security links to SentinelOne's product page.","cited_cells":8,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/prompt-security-sentinelone--sentinelone-com"},{"id":"promptlayer--promptlayer-com","name":"PromptLayer","url":"https://www.promptlayer.com/","section":"Evaluation & observability platforms","tier":3,"status":"active","last_verified_at":"2026-10-27","type":"Prompt management + tracing","license":"proprietary; Python SDK Apache-2.0","latest_release":"v1.5.16 (2026-08-19) — promptlayer","deployment":"Both (self-hosted + managed cloud) — self-hosting on Enterprise plan (GCP, AWS, Azure)","pricing":"Free $0 (2.5k requests/mo); Pro $49/mo; Team $500/mo (100k+ requests); Enterprise custom","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"MagnivOrg/prompt-layer-library — ~0.8k stars, last commit 2026-08","description":"Prompt versioning + tracing platform — non-eng-friendly prompt editor; early entrant pre-LLM-ops wave.","cited_cells":20,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/promptlayer--promptlayer-com"},{"id":"pydantic-logfire--pydantic-dev","name":"Pydantic Logfire","url":"https://pydantic.dev/logfire","section":"Evaluation & observability platforms","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary; Python SDK MIT","latest_release":"last commit 2026-10-29","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/pydantic/logfire (4526★)","description":"OpenTelemetry-based AI observability platform that traces model calls, agent steps, API requests and database queries in one view. SDKs are available for Python, JS/TS and Rust.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/pydantic-logfire--pydantic-dev"},{"id":"pyrit--microsoft-github-io","name":"PyRIT","url":"https://microsoft.github.io/PyRIT/","section":"Guardrails & AI security","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Framework","license":"MIT","latest_release":"last commit 2026-10-29","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/microsoft/PyRIT (4712★)","description":"Microsoft's Python Risk Identification Tool, an open framework for automated and human-led red teaming of generative AI systems. It runs multi-turn attack strategies such as Crescendo and TAP against any model or HTTP target.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/pyrit--microsoft-github-io"},{"id":"qlora--arxiv-2305-14314","name":"QLoRA","url":"https://arxiv.org/abs/2305.14314","section":"Training infrastructure","tier":3,"status":"stale","last_verified_at":"2026-10-27","type":"Quantized LoRA","license":"MIT — artidoro/qlora (Guanaco weights follow LLaMA license)","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"artidoro/qlora — ~11.1k stars, last commit 2024-06","description":"4-bit quantized LoRA fine-tuning — train 65B models on a single 48GB GPU. UW (Tim Dettmers).","cited_cells":12,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/qlora--arxiv-2305-14314"},{"id":"ragas--gh-explodinggradients-ragas","name":"Ragas","url":"https://github.com/vibrantlabsai/ragas","section":"Evaluation & observability platforms","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"RAG eval metrics","license":"Apache-2.0","latest_release":"v0.4.3 (2026-01-13) — ragas","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"vibrantlabsai/ragas — ~16.2k stars, last commit 2026-02","description":"Open-source eval framework for RAG pipelines — faithfulness, answer-relevance, context-precision, context-recall. Widely used as a baseline RAG eval kit.","cited_cells":18,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ragas--gh-explodinggradients-ragas"},{"id":"ray-train--docs-ray-io","name":"Ray Train","url":"https://docs.ray.io/en/latest/train/train.html","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Distributed training on Ray","license":"Apache-2.0","latest_release":"v2.59.0 (2026-10-02) — ray","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"ray-project/ray — ~44.1k stars, last commit 2026-10","description":"Ray's distributed training library — multi-node PyTorch/TF jobs; integrates Ray Tune and Ray Serve. Used at scale by OpenAI/Uber/Anyscale.","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ray-train--docs-ray-io"},{"id":"reka-core-flash-edge--reka-ai","name":"Reka Core / Flash / Edge","url":"https://www.reka.ai/","section":"Foundation models (substrate reference)","tier":2,"status":"active","last_verified_at":"2026-10-29","type":"Multimodal foundation model family — Reka Core / Flash / Edge","license":"Reka Flash 3 weights under Apache 2.0; Core proprietary","latest_release":"Reka Edge 2603 (2026-03); Rho-1 research preview (2026-10-05)","deployment":"Both (self-hosted + managed cloud) — Reka API (developer.reka.ai, also serves third-party open models); Reka Flash 3 weights on Hugging Face","pricing":"Reka API pay-per-token, per-model rates (e.g. Reka Edge 2603 $0.10 in / $0.10 out per 1M)","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"single-vendor (Reka / Snowflake)","github":"github.com/reka-ai (limited public repos)","description":"SF / Singapore foundation-model startup founded by ex-DeepMind / Meta / Google researchers (Dani Yogatama, Yi Tay, Qi Liu). Native multimodal models: Reka Core (~67B, frontier candidate), Flash (~21B), Edge (~7B). Apr-2024: claimed Core matched GPT-4 / Gemini Ultra on MMMU. Remains independent (a reported 2025 Snowflake deal did not happen). In 2026 it repositioned as a physical AI research lab (omni and world models, Reka Edge, Reka Cloud) and merged in a Moonvalley research team (2026-06-09).","cited_cells":36,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/reka-core-flash-edge--reka-ai"},{"id":"replicate--replicate-com","name":"Replicate","url":"https://replicate.com/","section":"Inference platforms & gateways","tier":1,"status":"acquired","last_verified_at":"2026-10-27","type":"Model-as-API inference cloud","license":"proprietary; Cog Apache-2.0","latest_release":"v0.23.1 (2026-10-20) — cog","deployment":"Managed-only — hosted API (Cog is OSS for packaging models)","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"replicate/cog — ~9.5k stars, last commit 2026-10","description":"Hosted-API platform for OSS models, part of Cloudflare since 2025-12-01 (still runs as a distinct brand) — Cog format packages models with their environment; thousands of community models behind a single REST API.","cited_cells":21,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/replicate--replicate-com"},{"id":"rl4lms--gh-allenai-rl4lms","name":"RL4LMs","url":"https://github.com/allenai/RL4LMs","section":"Training infrastructure","tier":3,"status":"research-code","last_verified_at":"2026-10-27","type":"AllenAI RL-for-LM library","license":"Apache-2.0","latest_release":"no releases","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"allenai/RL4LMs — ~2.4k stars, last commit 2024-03","description":"Allen AI's research library for RL training of language models — NLPO algorithm, NLP-task RL benchmarks; predates the TRL/OpenRLHF surge.","cited_cells":13,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rl4lms--gh-allenai-rl4lms"},{"id":"rlhflow--gh-rlhflow-rlhf-reward-modeling","name":"RLHFlow","url":"https://github.com/RLHFlow/RLHF-Reward-Modeling","section":"Training infrastructure","tier":4,"status":"stale","last_verified_at":"2026-10-27","type":"RLHF reward modelling toolkit","license":"Apache-2.0","latest_release":"no releases","deployment":"Self-hosted only — research codebase (training recipes)","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"RLHFlow/RLHF-Reward-Modeling — ~1.5k stars, last commit 2025-04","description":"Research codebase for RLHF reward modelling — pairwise / Bradley-Terry / pointwise reward heads; release of RewardBench-winning models.","cited_cells":13,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rlhflow--gh-rlhflow-rlhf-reward-modeling"},{"id":"roboflow--roboflow-com","name":"Roboflow","url":"https://roboflow.com/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Computer-vision data platform","license":"proprietary; Python SDK Apache-2.0","latest_release":"v1.7.3 (2026-10-26) — roboflow","deployment":"Both (self-hosted + managed cloud) — hosted training/inference; self-hosted Roboflow Inference free on own hardware; on-prem and air-gapped on Enterprise","pricing":"Free $0 (10 credits/mo); Core $39/mo (credit packs from $89/mo); Enterprise custom","mcp_support":"yes — official Roboflow MCP server and Skills installable into coding agents (Claude, Codex, Cursor, Gemini CLI)","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"roboflow/inference — ~2.5k stars, last commit 2026-10","description":"End-to-end computer-vision platform — dataset hosting, annotation, training, deployment. Universe public-dataset library; widely used for production CV.","cited_cells":20,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/roboflow--roboflow-com"},{"id":"rsl-rl--gh-leggedrobotics-rsl-rl","name":"RSL-RL","url":"https://github.com/leggedrobotics/rsl_rl","section":"Training infrastructure","tier":3,"status":"active","last_verified_at":"2026-10-27","type":"Fast RL for robotics","license":"BSD-3-Clause","latest_release":"v5.5.1 (2026-09-09) — rsl-rl-lib","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"leggedrobotics/rsl_rl — ~3.1k stars, last commit 2026-10","description":"ETH RSL's lightweight PyTorch RL library — optimised for GPU rollouts in robotics simulation (Isaac/MuJoCo). Used in many legged-robot RL papers.","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rsl-rl--gh-leggedrobotics-rsl-rl"},{"id":"runpod--runpod-io","name":"RunPod","url":"https://www.runpod.io/","section":"Inference platforms & gateways","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"GPU cloud for AI","license":"proprietary; Python SDK MIT","latest_release":"v1.12.0 (2026-08-10) — runpod (Python SDK)","deployment":"Managed-only — GPU cloud (Pods, Serverless, Clusters)","pricing":"Pay-as-you-go per-second or per-hour GPU billing; e.g. H100 PCIe Pod $2.89/hr, H100 SXM Pod $3.99/hr (Secure Cloud), Serverless H100 $4.79/hr; reserved capacity via sales","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"GPU-cloud platform — Pods (long-running), Serverless (per-request), and Hub (community templates). Popular for fine-tuning and inference of OSS models.","cited_cells":17,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/runpod--runpod-io"},{"id":"sanctuary-ai--sanctuary-ai","name":"Sanctuary AI","url":"https://sanctuary.ai/","section":"Robotics foundation models & agent stacks","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"General-purpose humanoid (Phoenix) — cognitive-architecture-led","license":"","latest_release":"Phoenix Gen 7 (2023)","deployment":"Direct customer pilot","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"Internal Sanctuary cloud (Carbon arch.)","llm_lock":"","github":"","description":"Vancouver-based Physical AI company, founded 2018; originally built the Phoenix humanoid, and since 2026-06 deploys hardware-agnostic Physical AI on existing commercial industrial robots (Tier 1 automotive proof-of-concept), with hydraulic hands in development.","cited_cells":32,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sanctuary-ai--sanctuary-ai"},{"id":"scale-ai--scale-com","name":"Scale AI","url":"https://scale.com/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Data labelling at frontier scale","license":"proprietary; Python SDK MIT","latest_release":"v2.18.4 (2026-04-24) — scaleapi (Python SDK)","deployment":"searched not found","pricing":"No public pricing — contact sales","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"searched not found","description":"Data-labeling powerhouse for AI training — RLHF data, fine-tuning corpora, frontier-lab partnerships. Meta made a strategic investment in 2025-06 for a minority stake (Scale's own post gives a valuation above $29B, not the amount); founder Alexandr Wang joined Meta and stays on Scale's board.","cited_cells":15,"third_party_cells":5,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/scale-ai--scale-com"},{"id":"sglang--gh-sgl-project-sglang","name":"SGLang","url":"https://github.com/sgl-project/sglang","section":"Inference platforms & gateways","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Structured LLM-program runtime","license":"Apache-2.0","latest_release":"v0.5.22 (2026-10-16) — sglang","deployment":"Self-hosted only — LLM serving framework","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"sgl-project/sglang — ~38.3k stars, last commit 2026-10","description":"Fast LLM serving framework with RadixAttention (cache reuse across requests with shared prefixes) + structured programming primitives. UC Berkeley.","cited_cells":19,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sglang--gh-sgl-project-sglang"},{"id":"snorkel-ai--snorkel-ai","name":"Snorkel AI","url":"https://snorkel.ai/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Programmatic data labelling","license":"proprietary; Snorkel OSS library Apache-2.0","latest_release":"v0.10.0 (2024-02-27) — snorkel (OSS library)","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"snorkel-team/snorkel — ~6.0k stars, last commit 2026-10","description":"AI data company spun out of Stanford Snorkel weak-supervision research; now sells expert-built training datasets, RL environments, benchmarks and evals to frontier AI labs (Snorkel Data Series, custom data development).","cited_cells":15,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/snorkel-ai--snorkel-ai"},{"id":"stable-baselines3--gh-dlr-rm-stable-baselines3","name":"Stable-Baselines3","url":"https://github.com/DLR-RM/stable-baselines3","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Reliable RL algorithm implementations","license":"MIT","latest_release":"v2.9.2 (2026-10-26) — stable-baselines3","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"DLR-RM/stable-baselines3 — ~13.9k stars, last commit 2026-10","description":"PyTorch implementations of RL algorithms (PPO, SAC, DQN, etc.) — research-grade reliability; widely used as a reference benchmark.","cited_cells":17,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/stable-baselines3--gh-dlr-rm-stable-baselines3"},{"id":"surge-ai--surgehq-ai","name":"Surge AI","url":"https://www.surgehq.ai/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"Premium RLHF data service","license":"searched not found","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"searched not found","description":"Premium human-labelling company specialised in RLHF data and difficult expert tasks — Surge prides itself on bootstrapped, no-VC; major frontier-lab supplier.","cited_cells":10,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/surge-ai--surgehq-ai"},{"id":"tensorrt-llm--gh-nvidia-tensorrt-llm","name":"TensorRT-LLM","url":"https://github.com/NVIDIA/TensorRT-LLM","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"NVIDIA inference compiler for LLMs","license":"Apache-2.0","latest_release":"v1.2.1 (2026-04-20) — tensorrt-llm; GitHub pre-release v1.4.0rc1 (2026-10-22)","deployment":"Self-hosted only — inference library/runtime for NVIDIA GPUs","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"NVIDIA/TensorRT-LLM — ~15.0k stars, last commit 2026-10","description":"NVIDIA's TensorRT extension for LLM inference — compiles transformers to optimised kernels for Hopper/Blackwell. Used in NIM and downstream deployments.","cited_cells":17,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/tensorrt-llm--gh-nvidia-tensorrt-llm"},{"id":"text-generation-inference-tgi--gh-huggingface-text-generation-inference","name":"Text Generation Inference (TGI)","url":"https://github.com/huggingface/text-generation-inference","section":"Inference platforms & gateways","tier":2,"status":"archived","last_verified_at":"2026-10-29","type":"HF inference server","license":"Apache-2.0","latest_release":"v3.3.7 (2025-12-19)","deployment":"Self-hosted only — inference server (Docker)","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"huggingface/text-generation-inference — ~10.9k stars, archived 2026-03-21, last commit 2026-03","description":"Hugging Face's LLM inference server (continuous batching, FlashAttention, quantization). Entered maintenance mode, then archived 2026-03-21; HF recommends vLLM, SGLang, llama.cpp or MLX instead.","cited_cells":17,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/text-generation-inference-tgi--gh-huggingface-text-generation-inference"},{"id":"together-ai--together-ai","name":"Together AI","url":"https://www.together.ai/","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Inference + fine-tuning cloud","license":"proprietary; Python SDK Apache-2.0","latest_release":"v3.0.0 (2026-10-22) — together (Python SDK)","deployment":"Managed-only — serverless and dedicated inference, fine-tuning, GPU clusters","pricing":"Serverless pay-per-token (e.g. $0.15–$3.00 per 1M input tokens); dedicated endpoints per GPU-hour (H100 $5.49/hr); fine-tuning per 1M training tokens; GPU clusters from $3.19–$3.99 per H100-hour","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Cloud platform for OSS LLM inference and fine-tuning — serverless inference, dedicated endpoints, fine-tuning API, training cluster.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/together-ai--together-ai"},{"id":"torchtune-pytorch--gh-pytorch-torchtune","name":"torchtune (PyTorch)","url":"https://github.com/meta-pytorch/torchtune","section":"Training infrastructure","tier":2,"status":"archived","last_verified_at":"2026-10-29","type":"Native PyTorch LLM fine-tuning","license":"BSD-3-Clause","latest_release":"v0.6.1 (2025-04-07) — torchtune","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"meta-pytorch/torchtune (linked from pytorch/torchtune) — ~5.8k stars, last commit 2026-04","description":"Pure-PyTorch library for fine-tuning LLMs — Llama/Mistral/Phi/Gemma recipes; native FSDP2 + memory-efficient training. From Meta PyTorch team. Per its README, torchtune is no longer actively maintained; active development ended in 2025.","cited_cells":15,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/torchtune-pytorch--gh-pytorch-torchtune"},{"id":"toyota-research-institute-lbm--tri-global","name":"Toyota Research Institute LBM","url":"https://www.tri.global/our-work/robotics","section":"Robotics foundation models & agent stacks","tier":1,"status":"active","last_verified_at":"2026-10-29","type":"Robotics research lab — Large Behavior Models (LBM)","license":"Some research code OSS (BSD / MIT)","latest_release":"LBM paper, A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation (2025-07-07)","deployment":"Research-stage; Toyota internal pilots; OSS code for community","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"Internal TRI / Toyota data centers","llm_lock":"","github":"ToyotaResearchInstitute (org) — 123 public repos incl. lbm_eval (~125 stars), last activity 2026-10","description":"Toyota Research Institute's robotics group — published Large Behavior Models (LBMs) Sep-2023 + diffusion-policy / behavior-cloning at scale. Industrial-research lab with Toyota's manufacturing-data advantage. Not a startup — division of Toyota Motor Corp.","cited_cells":35,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/toyota-research-institute-lbm--tri-global"},{"id":"trl-hugging-face--gh-huggingface-trl","name":"TRL (Hugging Face)","url":"https://github.com/huggingface/trl","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Transformer Reinforcement Learning library","license":"Apache-2.0","latest_release":"v1.16.0 (2026-10-15) — trl","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"huggingface/trl — ~19.6k stars, last commit 2026-10","description":"Hugging Face's open-source library for fine-tuning and aligning transformer models with RLHF, DPO, ORPO, KTO and PPO trainers. The default RLHF stack on top of HF Transformers.","cited_cells":18,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/trl-hugging-face--gh-huggingface-trl"},{"id":"truefoundry-ai-gateway--truefoundry-com","name":"TrueFoundry AI Gateway","url":"https://www.truefoundry.com/ai-gateway","section":"Inference platforms & gateways","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Enterprise AI gateway providing unified access to 1,600+ models with policy control, failover, routing, guardrails, budgets and monitoring. It can be deployed in VPC, on-prem or air-gapped environments.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/truefoundry-ai-gateway--truefoundry-com"},{"id":"truera--truera-com","name":"TruEra","url":"https://truera.com/","section":"Evaluation & observability platforms","tier":2,"status":"acquired","last_verified_at":"2026-10-29","type":"ML observability + RAG triad","license":"searched not found","latest_release":"v12.6.1 (2024-03-04) — truera (Python SDK)","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"ML observability with the TruLens open-source LLM eval framework — the 'RAG Triad' (context-relevance, groundedness, answer-relevance) metrics. Acquired by Snowflake 2024-05.","cited_cells":9,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/truera--truera-com"},{"id":"trulens--gh-truera-trulens","name":"TruLens","url":"https://github.com/truera/trulens","section":"Evaluation & observability platforms","tier":2,"status":"active","last_verified_at":"2026-10-29","type":"LLM eval framework","license":"MIT","latest_release":"v2.16.0 (2026-10-19) — trulens","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"yes — instruments MCP tool calls via the MCP span type (tool name, arguments, output, latency)","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"truera/trulens — ~3.6k stars, last commit 2026-10","description":"Open-source LLM/RAG evaluation framework — feedback functions, the RAG Triad. Originally TruEra; now Snowflake-stewarded.","cited_cells":20,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/trulens--gh-truera-trulens"},{"id":"unsloth--unsloth-ai","name":"Unsloth","url":"https://unsloth.ai/","section":"Training infrastructure","tier":1,"status":"active","last_verified_at":"2026-10-27","type":"Fast/efficient LLM fine-tuning","license":"Apache-2.0 (free tier)","latest_release":"v2026.10.9 (2026-10-26) — unsloth","deployment":"Self-hosted only — library plus local Unsloth Studio web UI","pricing":"Free OSS; Pro/Enterprise paid","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"unslothai/unsloth — ~78.6k stars, last commit 2026-10","description":"Fast and memory-efficient LLM fine-tuning library — claims 2-5× speedup, 60-80% memory savings on a single GPU vs HF baseline through custom Triton kernels.","cited_cells":21,"third_party_cells":5,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/unsloth--unsloth-ai"},{"id":"vellum-ai--vellum-ai","name":"Vellum AI","url":"https://www.vellum.ai/","section":"Evaluation & observability platforms","tier":2,"status":"dead","last_verified_at":"2026-10-29","type":"LLM dev platform (build + eval + deploy)","license":"proprietary; Python SDK MIT","latest_release":"v1.14.8 (2026-10-02) — vellum-ai (Python SDK)","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"searched not found","description":"Originally an enterprise LLM development platform (prompts, workflows, evals). It pivoted to a personal AI assistant (open-source vellum-assistant plus Vellum Cloud). Vellum announced its hosted Personal Assistant Platform will be shut down on 2026-11-02 and that it will no longer actively maintain the open-source repo.","cited_cells":13,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/vellum-ai--vellum-ai"},{"id":"vercel-ai-gateway--vercel-com","name":"Vercel AI Gateway","url":"https://vercel.com/docs/ai-gateway","section":"Inference platforms & gateways","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Product","license":"proprietary","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","description":"Vercel's unified endpoint for text, image, video and speech models across providers, used from the AI SDK and OpenAI-compatible clients.","cited_cells":11,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/vercel-ai-gateway--vercel-com"},{"id":"verl--gh-volcengine-verl","name":"Verl","url":"https://github.com/verl-project/verl","section":"Training infrastructure","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Volcano RLHF framework","license":"Apache-2.0","latest_release":"v0.9.1 (2026-09-20) — verl","deployment":"Self-hosted only — library","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"verl-project/verl (formerly volcengine/verl) — ~24.0k stars, last commit 2026-10","description":"ByteDance Volcano-Engine's open RLHF framework — flexible HybridFlow architecture; supports PPO/GRPO at large scale.","cited_cells":14,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/verl--gh-volcengine-verl"},{"id":"vllm--gh-vllm-project-vllm","name":"vLLM","url":"https://github.com/vllm-project/vllm","section":"Inference platforms & gateways","tier":1,"status":"active","last_verified_at":"2026-10-26","type":"Fast LLM inference engine","license":"Apache-2.0","latest_release":"v0.32.0 (2026-10-18)","deployment":"Self-hosted only — pip / Docker inference server with OpenAI-compatible API; managed hosting only via third-party clouds","pricing":"Free (open source, Apache-2.0)","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"","llm_lock":"","github":"vllm-project/vllm — ~95.4k stars, last commit 2026-10","description":"Open-source LLM inference engine with PagedAttention — high-throughput batching, paged KV-cache. UC Berkeley origin; de-facto OSS inference stack.","cited_cells":19,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/vllm--gh-vllm-project-vllm"},{"id":"weights-biases--wandb-ai","name":"Weights & Biases","url":"https://forge.coreweave.com/wandb","section":"Training infrastructure","tier":1,"status":"acquired","last_verified_at":"2026-10-29","type":"ML experiment tracking + ops","license":"proprietary; wandb SDK MIT","latest_release":"v0.31.0 (2026-10-16) — wandb","deployment":"searched not found","pricing":"CoreWeave Forge plans: Free $0 (up to 5 model seats); Pro from $60/mo; Enterprise custom; Academic free","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"wandb/wandb — ~11.3k stars, last commit 2026-10","description":"Experiment-tracking and ML-ops platform (metrics, artifacts, sweeps, Weave tracing/evals). Acquired by CoreWeave (completed 2025-05-05, ~$1.7B); now part of CoreWeave Forge.","cited_cells":17,"third_party_cells":7,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/weights-biases--wandb-ai"},{"id":"xai-grok-4--x-ai","name":"xAI Grok 4","url":"https://x.ai/news/grok-4","section":"Foundation models (substrate reference)","tier":1,"status":"active","last_verified_at":"2026-10-31","type":"Frontier foundation model from xAI (Grok 4)","license":"proprietary (Grok 4); older open weights: Grok-1 (Apache-2.0) and Grok 2 (Grok 2 Community License)","latest_release":"Grok 4.7 (2026-09-21)","deployment":"Managed-only — xAI API, grok.com, X apps; also via Amazon Bedrock, Microsoft Foundry, GitHub Copilot (older Grok-1 / Grok 2 weights downloadable)","pricing":"API: Grok 4.7 $2 in / $6 out per 1M (prompts ‹200k tokens; $4 / $12 above); consumer via X / SuperGrok subscriptions","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"single-vendor (xAI-only family)","github":"github.com/xai-org/grok-1 ~50k stars (older Grok 1 weights only)","description":"Elon Musk's xAI frontier model — Grok 4 launched 2025-07. Trained on Colossus (100k+ GPU supercluster in Memphis). Deeply integrated with X/Twitter (real-time tweet data access). Grok 4 Heavy variant for premium tier. No frontier-tier open-weights — Grok 1 (314B MoE) is the only weight release. The x.ai site now brands itself SpaceXSI (page title, checked 2026-10-29).","cited_cells":37,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/xai-grok-4--x-ai"},{"id":"zenml--zenml-io","name":"ZenML","url":"https://zenml.io/","section":"Training infrastructure","tier":2,"status":"active","last_verified_at":"2026-10-27","type":"ML pipeline orchestration","license":"Apache-2.0","latest_release":"v0.98.0 (2026-10-14) — zenml","deployment":"Both (self-hosted + managed cloud) — OSS self-hosted; ZenML Pro as SaaS or self-hosted/air-gapped on Enterprise","pricing":"Open Source free (self-hosted); Scale $999/mo per workspace (SaaS); Enterprise custom","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"zenml-io/zenml — ~5.6k stars, last commit 2026-10","description":"Open-source ML pipeline orchestration framework — connects sklearn/PyTorch/HF pipelines to MLflow/W&B/cloud infra. Pivoting toward LLM-ops.","cited_cells":18,"third_party_cells":3,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/zenml--zenml-io"}]}