Open-source agent memory systems

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

Acquired by ClickHouse (2026-01-16); product and MIT-licensed repo continue under the Langfuse name. Memory module reads/writes captured as named spans. Trace Log View concatenates every agent step including memory ops. Agent Graphs (GA …

Modular graph-based RAG. Leiden community detection over knowledge graph; community summaries at multiple levels. DRIFT Search hybrid (Oct 2024). The GraphRAG repo is now largely in maintenance mode: no new features or PRs, only bug fixe…

Bi-temporal knowledge graph (event time + ingestion time). Strong on chronological reasoning and contradiction tracking. Graphiti is the open-source core.

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

Discontinued: Flowise (acquired by Workday, Aug 2025) was sunset in 2026 — code freeze 2026-07-29, GitHub repo archived 2026-08-13, end of life 2026-08-31. Was a visual LangChain canvas with LangChain-native memory nodes (Buffer, Buffer …

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

Originated as MemGPT (UC Berkeley), where the model pages its own memory between context, recall and archival tiers. The product now centres on Letta Code, a stateful agent harness whose agents rewrite their own memory, skills and prompt…

Knowledge-graph-based persistent memory MCP server — the canonical reference implementation. npx -y @modelcontextprotocol/server-memory ; configurable via MEMORY_FILE_PATH .

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.

Cloud-free agent memory using information-geometric retrieval (Fisher-Rao distance + BM25 + entity graph + temporal fusion via 4-channel RRF) instead of LLM calls for scoring and lifecycle. Mode A is the only system to cross 70% LoCoMo w…

Open Markdown 'README for agents' at repo root; nested files in subprojects take precedence (closest wins). Holds build/test commands, conventions and PR rules. Stewarded by the Agentic AI Foundation (Linux Foundation); supported by Code…

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.

Hooks into Claude Code's session lifecycle; compresses tool outputs into ~500-token observations via Claude's agent SDK. SQLite + FTS5 with optional Chroma vector. "Endless Mode" compresses in-flight to extend session length. Captures wh…

High-performance MCP server building persistent code knowledge graphs via AST. 155 languages via vendored tree-sitter; SQLite + WAL; single static binary; zero runtime deps. Knowledge graph: Project/Package/File/Module/Class/Function/Met…

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."

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.

Archived 2026-03-24 (read-only); README now points to Mem0's official cloud-hosted MCP server at mcp.mem0.ai/mcp. Mem0's managed cloud API exposed as MCP tools. Semantic search + CRUD on memories scoped by user / agent / session. Cloud-h…

Commercial Mem0 ships SOC 2 / HIPAA, zero-trust access controls, BYOK encryption, real-time monitoring, audit logs and workspace governance as defaults (Trust Center at trust.mem0.ai). For privacy-first deployments the self-hosted path i…

Vibe is Mistral's cloud-hosted coding-agent platform: each session runs in an isolated sandbox with task state, history, and approvals. Spawnable from CLI or Le Chat, with parallel execution. Local CLI sessions can be "teleported" to the…

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…

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…

Extends π0 with MEM layer giving the policy short-term (within-task) + long-term (cross-task, >10-min horizon) memory. Co-trained on robot teleoperation + human video + text. Hierarchical inference: VLA predicts subtasks as text tokens, …

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…

When Mem0 is connected, gains Memory Operation Timeline, Search Analytics, Memory Growth tracking, Error Tracking per memory call. Standalone, records context at each step but doesn't analyse memory quality.

AutoGen's memory layer: ListMemory chronological context plus teachable agents that vectorise corrections; integrates with Mem0/Zep rather than building deep memory natively. Microsoft AutoGen is in maintenance mode (community managed, b…

Structured, persistent memory layer for AI agents. Open source.

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).

Appends conversation turns to .aider.chat.history.md in project root. restore_chat_history defaults to false — no automatic replay into new sessions. Community /session save command for JSON state snapshots. .aider.conf.yml carries stati…

Memory subsystem inside the CrewAI orchestration framework; integrates with Mem0 for the long-term tier.

User-profile-based long-term memory. Per-user profile + event timeline. API + Python/Node/Go SDKs + MCP server. Profile-centric rather than fact-centric.

Microsoft Semantic Kernel's memory plugin (Microsoft.SemanticKernel.Plugins.Memory): a plug-in approach to long-term memory in the SK agent framework (agent memory marked experimental). Semantic Kernel's README names Microsoft Agent Fram…

Open-source AI automation platform. Routes between short-term and long-term memory adaptively across any LLM provider; plugin system for storage backends. Memory managed at the instruction-management layer — task context, instruction sta…

Agno (formerly Phidata). AgentStorage persists sessions to a DB; AgentMemory auto-classifies/store user preferences and conversation summaries. Single-line integrations with LanceDB, Pinecone, Weaviate, Qdrant.

Renamed Anda Brain (repo now ldclabs/anda-brain). Native graph-based memory for autonomous AI agents. Bio-inspired sleep-based memory consolidation. The hosted service brain.anda.ai has been discontinued; Anda Brain is now self-hosted on…

Tsinghua spinout. Foundation agent series for autonomous smartphone/PC GUI control. Persistent task memory retains shopping history, app usage patterns, multi-step task state across sessions. AutoGLM-Thinking claimed first agent capable …

Stateful AI API combining persistent memory, RAG, web search and routing across many LLMs behind one interface, with SDKs for Python and npm; benchmarks its memory on LoCoMo and LongMemEval. As of Oct 2026 Backboard positions itself as a…

In-browser WebContainer (Rust-backed shared memory) is the implicit state store. No dedicated memory abstraction — project files are the state. Team plan adds shared workspaces + GitHub integration.

Claude Code plugin packaging of Memvid (see Dedicated memory layers). Wraps the Memvid .mv2 single-file store as a slash-command-installable plugin so per-project memory becomes a git-committable file inside the repo.

FastAPI server. RBAC + API-key auth, encryption, backup/recovery, embedding pipeline with deduplication. Only server in this list with role-based access control over pgvector.

Wraps Supermemory cloud API. Team memory shared across project members, separate from personal memories. Auto-capture on session end; per-repo config.

"Cognify" extracts entities and relationships into a structured graph. Saves interaction patterns as coding rules; uses both vector and graph DBs. Bundled in Cognee v0.3.5+.

.continue/rules/*.md at workspace root, loaded lexicographically; YAML frontmatter name, globs, regex, description, alwaysApply. Continue was acquired by Cursor in June 2026 and the extension is no longer maintained.

SQLite + FTS5; agent decides what to save (not firehose capture). Single Go binary, no CGO, no Node/Python/Docker dependencies. HTTP API + MCP + CLI + TUI; optional cloud sync with local-source-of-truth.

Google's Agent Development Kit. Session services: InMemory / Database / VertexAI. Long-term: InMemoryMemoryService or VertexAIMemoryBankService (uses Vertex AI Memory Bank Preview for persistent cross-session memory).

World (facts), Experiences (agent actions), Opinion (beliefs with confidence scores), Observation (mental models). Entity resolution ("Alice" and "my coworker Alice" same node). Mimics human recall topology.

First-class memory module inside LangChain — explicit cognitive-science-style typing of memories.

Two-tier model: ChatMessage FIFO queue with overflow flushed into long-term Memory Block objects. Memory class replacing older ChatMemoryBuffer.

SQLite + FTS5 + BM25 ranking. Stores learnings, decisions with reasoning and alternatives , entities, observations. Loads context from last 3 sessions. Only server that captures why a decision was made + alternatives considered.

Open-source local-first outliner. Notes stored as Markdown / Org-mode on disk. AI integration via community plugins (e.g. logseq-ai for chatting with the graph). Graph view + backlinks; daily-journal-driven workflow.

ByteDance Seed + Zhejiang Univ + SJTU. Multimodal agent with long-term memory as entity-centric graph built from live video + audio streams. Parallel memorisation + control processes. Ships M3-Bench evaluation dataset.

TypeScript-first agent framework. Default memory: LibSQL store + on-device fastembed-js for semantic recall.

KG-based local fork of @modelcontextprotocol/server-memory . Avoids ephemeral npx installs; supports named databases for topic scoping; user-controlled memory path. Drop-in for teams that want the official KG model without npx churn.

Embeddings via @cf/baai/bge-m3 ; Vectorize + D1; per-user namespace isolation. Runs entirely on Cloudflare Workers — no separate server.

SQLite-Vec default, Cloudflare hybrid for production. Typed edges (causes, fixes, contradicts). Autonomous consolidation compresses old memories. SSE real-time notifications + D3.js visualisation dashboard. Agent-pipeline focused (LangGr…

Open-source long-term memory framework for autonomous agents.

Multi-layered open-source memory system for AI agents.

Shenzhen. Argues attaching external vector stores is architecturally flawed; trains memory as an intrinsic capability via Agentic RL. Plans B2B enterprise KM and C2C personalisation in 12 months. Founded by Tianjin University AI professo…

Memory layer from MatrixOrigin. Strong-retrieval / clear-reader-separation design.

Walrus is a Sui-blockchain decentralised storage protocol; MemWal is its AI-facing SDK that stores encrypted agent memory on-chain — portable, verifiable, shareable across agent systems with no single-vendor lock-in. Targets production a…

Cross-Model Verification Kernel (CMVK) requires majority-voting agreement across multiple model calls before a memory-influenced action. Agent OS package intercepts every action (memory included) at sub-millisecond latency. MIT licensed.

Renamed NVIDIA CMX Context Memory Storage Platform (2026). A BlueField-4 (STX) powered "AI-native context tier" that extends GPU memory with a shared pod-level KV-cache tier for long-context, multi-turn agentic inference, with Dynamo rou…

Omni memory system for personalised, long-horizon, self-evolving agents. From OPPO PersonalAI.

Local context/memory engine for coding agents: stores decisions, docs and agent history on-machine (SQLite + local ONNX embeddings) and serves a ranked slice per task over MCP. Open-core Apache-2.0 core with paid Pro tier.

OpenAI's agent SDK. Sessions for working context within an agent loop; RunContextWrapper for structured persistent state. April 2026 update added categorised long-term memory: project / user / policy. Long-term tier typically backed by e…

Renamed LongMemory (repo CaviraOSS/LongMemory; packages longmemory / longmemory-sdk, with openmemory-js@2 / openmemory-py@2 as forwarding bridges). Open-source cognitive memory engine for LLMs and agents.

Mem0's MCP memory layer for coding agents (Claude, Cursor, VS Code, Windsurf, Codex etc.). As of Oct 2026 it is a hosted service: the MCP server runs at mcp.mem0.ai, needs a Mem0 Platform account (browser sign-in or API key, npx mcp-add …

Pydantic-AI ships no native persistent memory; agents start fresh each run. Long-term memory typically added via third-party integration (e.g. Hindsight) using tools + auto-injected memory instructions.

Only tool found that scans the persistent memory layer on disk rather than runtime tracing. Detects injected instructions, obfuscated payloads (zero-width Unicode, base64, homoglyphs, hex), C2-pattern URLs, credential leakage. ratine dif…

Daily-notes + backlinks note app now relaunched as Reflect Open: open-source, offline-first, notes stored as local Markdown files synced via iCloud or Git; optional AI with your own API key. Reflect Classic still maintained.

Discontinued: Roo Code (VS Code agent) used .roorules / .roo/rules files as persistent project instructions; the extension shut down on 2026-05-15 and its repo is archived.

Hugging Face barebones agent framework. Memory lives as a Python list; no built-in persistence. Step callbacks let you mutate memory between steps. Persistence is left to the user — community proposals exist for "memory bank"-style addit…

AWS code-first agent framework. Multi-LLM (Bedrock, Anthropic, Gemini, OpenAI). Memory via Mem0 integration (ElastiCache for Valkey + Neptune Analytics) or Bedrock AgentCore Memory. Local default uses FAISS; OpenSearch for production.

Claude Code plugin that indexes prior conversations at session end and exposes search/show tools via MCP. Distributed via obra/superpowers-marketplace ; install with /plugin install episodic-memory@superpowers-marketplace .

Superseded: the sweepai/sweep README says the team now builds an AI coding assistant for JetBrains; the GitHub-app product described here is no longer the focus (repo not archived). No dedicated persistent memory layer. Operates as a Git…

Wraps conversation memory behind an Advisor abstraction. MessageWindowChatMemory default 20-msg sliding window (preserves system messages); pluggable ChatMemoryRepository backends (in-memory, JDBC, Cassandra, Neo4j). Memory injected as c…

DeepMind. Retrieval-Enhanced Transformer with chunked cross-attention over a 2T-token retrieval database. Frozen BERT retriever + differentiable encoder.

Meta AI. Jointly fine-tunes retriever + LM for few-shot tasks. The facebookresearch/atlas repo is no longer maintained; the code is provided as is.

Stanford recursive-summarisation tree over the corpus — clusters and summarises documents into a hierarchical index; retrieves at multiple levels. ICLR 2024.

Hippocampus-inspired RAG. LLMs + knowledge graphs + Personalized PageRank to mimic neocortex/hippocampus roles. NeurIPS 2024.