Self-hostable agent memory systems

Universal memory layer for AI agents. Three concurrent stores (vector + graph + KV); LLM-extracted facts; concurrent retrieval via ThreadPoolExecutor.

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. Multimod…

Property graph DB with Cypher; native vector search; Aura Agent + MCP server for graph-as-memory. $100M GenAI investment in 2025.

Distribution-Based Score Fusion + RRF. Sparse vectors native; filtering via ANN graph modification.

Stores embeddings alongside relational + full-text data. HNSW + IVFFlat ANN indexes. Used as agent conversation memory via LangChain + MCP. Foundation of Supabase AI and many self-hosted RAG stacks.

Managed vector DB. Cascading sparse + dense + rerank pipeline; pinecone-rerank-v0 .

Limited native hybrid (users build RRF custom). Fast Rust core (v2.5).

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/Ch…

"Extract–Cognify–Load" pipeline that turns raw input into a typed, queryable knowledge graph for agent recall.

Standalone memory service from Vectorize. Open source. Biomimetic four-network design (World, Bank, Observation, Opinion). Ships an MCP memory server.

Embedded vector DB (Arrow columnar). RRF reranker. Petabyte-scale on disk.

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 fo…

Multi-vector columns (10 simultaneous). Native hybrid search (v2.5). CAGRA + Vamana GPU/CPU (v2.6).

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. Rais…

Memory engine with API, app, browser extension, and MCP server. Extracts facts, tracks updates, resolves contradictions, auto-forgets expired info. Plugins for Claude Code, OpenCode, OpenClaw, Hermes.

Hybrid Search 2.0 with learned ranking. BM25 + vector in unified index.

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 op…

Runtime enforcement targeting memory poisoning, unauthorised execution, identity spoofing per the OWASP LLM threat list. Visibility/control over MCP servers and locally-installed AI tools — the infrastructure layer where memory is most e…

German enterprise stack. PhariaCatch captures + preserves organisational knowledge through structured human-machine collaboration; PhariaAssistant queries via semantic search. Combines sovereign deployment constraints with explainability…

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, Q…

RDF triple/quad store with RDFS++ / OWL reasoning. v8.4 (May 2025) added an NLQ interface.

Folder + tag hierarchy; iCloud sync. Writing Tools (proofread / rewrite / summarise) across all note content; audio transcription + AI summary; smart search; reminders from to-dos.

HybridGraphRAG combines vector search, graph traversal, full-text in one AQL query. ArangoGraphML for ML pipelines; LangChain integration.

Real-time knowledge graph indexed from connected repos + commit history + CI/CD + tickets + docs. Separate Agent Memory layer (Sept 2025) captures session decisions/corrections/goals as discrete items with developer review/approval UI. M…

Per-plan vector-DB storage quota + LLMz autonomous engine (in-session working memory) + Knowledge Base (semantic search over uploaded docs). Long-term user memory persists across sessions.

Python library that turns any LLM into a browser agent — extracts DOM into a structured representation the LLM can act on (rather than pure screenshot). Browser Use Inc. raised $17M seed Q1 2025 (Felicis); cloud product at cloud.browser-…

Bibliographic metadata curation (Web of Science, Derwent, Cortellis). Human editorial governance + journal-deindexing. Memory-adjacent — included as a curated-knowledge baseline.

Enterprise contact centre (voice + chat). Short-term covers current session; long-term persists across sessions and is shared by every agent a customer might encounter. Most explicitly engineered memory model in the segment.

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 (smalle…

Mimir hierarchical memory: short-term (verbatim), medium-term (summaries + emotion), long-term (cross-session). Scene awareness layer adds in-engine perception. Unity + Unreal SDKs with real-time voice + lipsync.

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 d…

Structured Knowledge items stored server-side. Three-tier scope: repo-pinned (always injected), global (recall by relevance), org/enterprise (visible to org members). Auto-generates Repo Knowledge from repo structure each session; reads …

e2b provides secure, ephemeral sandboxes (Firecracker microVMs) for AI agent code execution. Each sandbox spins up in <200ms with full Linux environment, Python/Node/etc preinstalled. Substrate for code-agent products including Anthropic…

Mature hybrid (BM25 + vector) implementations. RRF + weighted combination. Elastic ELSER sparse model.

Applies guardrails at three points: (1) before write to vector DB, (2) before query reaches embedding model, (3) before response. Detects malicious instructions in stored memory before retrieval; scans for PII/PHI in/out. Text + image + …

Visual LangChain canvas. LangChain-native memory nodes (Buffer, Buffer Window, Conversation Summary Buffer) backed by internal chat_message DB table. Buffer Window limits to last K interactions; Summary Buffer compresses older turns whil…

Treats memory as first-class in multi-agent tracing. Luna-2 SLMs (3B / 8B) scan every interaction for intent drift and belief drift; 20+ checks at sub-200ms latency. Catches when agent A's view of the world splits from teammate B's. Open…

GA at Google Cloud Next 2026 (2026-04-22). Vertex AI rebranded as Gemini Enterprise Agent Platform; Memory Bank now "Agent Platform Memory Bank" with new continuous event-streaming + automated memory generation triggered by configurable …

Enterprise search with 100+ connectors. Personalised per-user knowledge graph. No governance layer.

Tracks how facts change over time; maintains provenance; supports prescribed and learned ontology. Bi-temporal edges (when known vs when valid). Only server that distinguishes "fact learned" from "fact true."

Personal lawyer memory + matter memory + institutional firm memory + client-institution memory. Opt-in at user level with admin kill-switches and ethical-wall controls. Announced Jan 2026.

Targets the supply-chain / lineage layer rather than runtime memory writes. Model Genealogy tracks training/fine-tuning/modification history — catches poisoning baked in during training. AIBOM generates auditable inventory of model compo…

Leading purpose-built NPC engine. Long-Term Memory (named feature) stores interaction facts across sessions; surfaces them contextually. Character Brain coordinates multiple ML models. Personality defined by trait axes, backstory, goals,…

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.

JetBrains' in-IDE assistant integrated across IntelliJ, PyCharm, GoLand, WebStorm, Rider, etc. Multi-model: OpenAI, Anthropic, Google, local. Differentiated by its deep IDE-API integration — refactorings, run configs, debugger context.

Spec-driven agentic IDE from AWS — preview launch July 2025. Codifies the AI-coding workflow as Requirements / Design / Tasks files generated before code ("specs"); also persists "steering" files and Hooks. Built as a VS Code fork; runs …

Screens every prompt, response, and retrieved document for indirect prompt injection — primary vector for memory poisoning. Treats memory as untrusted: anything written to or read from memory is adversarial until proven clean. Lakera Red…

Agent component has built-in chat memory enabled by default, keyed by session ID. Custom session IDs isolate per-user history. Bundled Mem0 + Redis integrations for external memory layers.

Memory module reads/writes captured as named spans. Trace Log View concatenates every agent step including memory ops. Agent Graphs (GA 2025) infer graph structure from observation nesting; session-level replay tracks how memory state ev…

LLM-as-Operating-System paradigm — model manages its own memory, paging between core context, recall buffer, and archival store. UC Berkeley origin; transitioning to letta_v1_agent .

LiveKit Agents — agent framework built on LiveKit's open-source WebRTC media server. LiveKit raised $45M Series B Sep-2024 ($300M val). Powers OpenAI Realtime API's WebRTC transport (LiveKit is the substrate OpenAI chose). Agents framewo…

Enterprise Q&A agent powered by Luci patent-pending ultra-precise RAG. Embeds on product detail pages; consumes technical PDFs, spec sheets, images, tables, charts, graphs and product manuals. Maintains session history for multi-turn fol…

Commercial Mem0 ships SOC 2 / HIPAA, zero-trust access controls, BYOK encryption, real-time monitoring, audit logs, workspace governance as defaults. OpenMemory is the local self-hosted variant (Docker + FastAPI + Postgres + Qdrant) for …

Windows-native screen-recording memory. Continuously indexes activity across apps, documents, browsing; timeline UI for revisiting work. Distinct from Copilot Memory.

Mistral's Agents API provides persistent conversation state across turns and sessions; runtime maintains branching conversation trees. Memory is implicit in the stateful session model rather than a separate store — agents can branch, con…

Le Chat Memories — graph-based, context-aware. Stores user preferences and past decisions; memories saved explicitly (user instruction) or inferred automatically. Coupled to "Work Mode" agentic layer.

Modal Labs ships a serverless container platform with first-class GPU support; Sandboxes feature (2024) adds ephemeral, AI-agent-friendly compute sandboxes. Founded 2021 by Erik Bernhardsson (ex-Spotify ML, author of Luigi + Annoy). $87M…

Low-code workflow platform. AI Agent node ships with pluggable memory sub-graph: Simple Memory for development; Postgres/Redis for production; vector store nodes (Qdrant, Pinecone, MongoDB Atlas) for semantic recall.

Wearable orb (necklace or clip) with open-source hardware + firmware. Continuous audio capture; real-time GPT-4o transcription; searchable memory graph. Supports app screen capture too. Hybrid privacy: cloud by default (SOC 2/HIPAA/AES-2…

RDF store with real-time inference. Ontotext merged with PoolParty in 2025 to form Graphwise.

Memory under ~/.codex/memories/ : summaries, durable entries, recent-input logs, supporting evidence — generated in background after sessions go idle. Chronicle (opt-in macOS preview) uses screen-capture to seed memory. Off by default; o…

OpenAI's open-source terminal coding agent — released April 2025 alongside o3/o4-mini. Counterpart to Claude Code in the OpenAI stack. Memory via ~/.codex/memories (file-backed); routes to OpenAI o-series and Codex (model) for execution.…

OpenAI's general-purpose browser agent — research preview launched January 2025 (US Pro tier). Built on the Computer-Using Agent (CUA) model that processes screenshots and emits mouse / keyboard actions; runs inside an OpenAI-hosted brow…

OpenAI's Realtime API for native speech-to-speech multimodal interaction with GPT-4o — launched Oct 2024 (DevDay). Skips the STT → LLM → TTS pipeline by passing speech directly to GPT-4o. The platform substrate that Vapi/Retell/Bland int…

Cross-listing of OpenHands at the harness layer. OSS Devin-alternative — sandboxed full SWE agent with browser + shell + editor. Renamed from OpenDevin Mar 2024. ~40k★. Commercial offering: All Hands AI cloud (managed SaaS).

Proprietary operational ontology — semantic objects + kinetic actions wired to microservices. Not open-standards KG; closed platform.

Open-source Python framework for real-time voice + multimodal AI agents, maintained by Daily.co. Pipeline-style architecture (STT → context → LLM → TTS) wired over WebRTC. The OSS substrate that many of the voice platforms build on or co…

Taxonomy and knowledge organisation. Taxonomy Advisor (LLM-powered). SKOS-based.

Hybrid retrieval combining dense vector + keyword + graph traversal + multimodal. Agentic RAG grounding AI agents in internal knowledge. SOC 2 / ISO 27001 compliant.

Sourcegraph's coding assistant — IDE plugin (VS Code, JetBrains) plus web. Distinguishes itself with Sourcegraph's code-search index as the context provider (cross-repo, codebase-wide retrieval). Agentic chat added 2024; runs Claude / GP…

RDF-native semantic platform with SHACL validation, virtual graphs, and the Voicebox AI assistant. $32.5M funding. Pitches itself as a governed-knowledge layer.

Parallel graph DB with GSQL/GQL. CoPilot for entity extraction. Free community edition.

Ontology-native semantic governance suite. SHACL + AI Copilots for taxonomy management.

Object-storage-first vector DB. Powers Notion and Cursor. SPFresh indexing; ~10× cost reduction vs. memory-resident peers.

Combines vector ANN, lexical, structured filtering, and ML-learned ranking in one distributed system. Integrated chunking + layered ranking for RAG. No separate embedding infra needed.

Codeium-built agentic IDE (VS Code fork); flagship was Cascade — a planning + execution agent that maintains "Flows" and "Memories" as durable artifacts. Reported acquired by OpenAI in May 2025 for ~$3B (later partly unwound — Google pai…

Native (Rust) collaborative code editor by Zed Industries (founded by the Atom + Tree-sitter team). Added Agentic Editing in 2024 — multi-step coding agent driven by Claude or local models; .rules markdown + Threads form the memory layer…

Bi-temporal knowledge graph preserves full lineage of every stored fact — every memory write traceable and auditable by graph schema, not bolted-on logging. SOC 2 Type II; HIPAA BAA available; BYOK + VPC-isolated deployment for data resi…

Distinct from LangMem. Built-in checkpointer saves graph state per super-step (short-term, thread-scoped). Store System adds long-term hierarchical key-value memory across threads with optional vector search + TTL. Postgres / Mongo / Red…

Single-file memory layer (one .mv2 file). No DB, no server. Append-only sequence of immutable Smart Frames with timestamps + checksums. Native Rust core (rewritten from Python).