Agent frameworks & infrastructure
Agent memory & context
4,032 agents · most starred first
Agent memory & context, page 25 of 41
Multi-user multi-agent memory system for OpenClaw. Replaces the default file-backed memory with an intelligent, layered architecture featuring ACT-R activation scoring, Ebbinghaus forgetting curves, Zettelkasten-style note linking, and automatic consolidation.
One shared Markdown file that AI agents take turns writing to — context, decisions and progress all live in the file, so any agent can pick up the work without you summarising the last chat. Single binary: CLI + MCP server + live Web UI
Official implementation of LifeMem: longitudinal memory for more faithful, diverse, and dynamically evolving LLM agents.
Research-oriented cognitive AI architecture exploring memory, reasoning, planning, and adaptive learning using local LLMs.
Local-first AI project memory and control runtime for evidence, privacy, model routing, human approvals and cross-harness handoffs.
Web API sample with Microsoft Agent Framework, Azure OpenAI, and SQL Server that adds persistent per-user memory to AI agents through authenticated chat endpoints and automatic memory extraction.
Previously on you. A self-evolving AI agent that remembers by when, not by chat thread — it learns who you are, out in the open. Memory as plain Markdown in your own repo.
Official Python and TypeScript SDKs for Anona Memory managed memory for AI agents. Record, retrieve and reason over memories, with framework adapters and an MCP server.
Local-first memory for AI coding agents that catches a mistake before it repeats. Plain Markdown + git, zero deps, every agent via MCP/hooks. Formerly Anamnesis.
Analyze OpenClaw session logs to generate weekly self-improvement suggestions for AI agents without silent modifications.
Agent-agnostic evidence-bound memory and handoff side-car for multi-agent systems. Claims carry checksummed excerpts — verified means re-verifiable, stale evidence gets flagged. MCP server, CLI, and git hooks. Apache-2.0.
ACE — Agentic Context Engineering: evolving, self-improving context playbooks for LLM agents. Faithful ICLR 2026 implementation with OpenAI Agents SDK support.
An intelligent memory gateway for AI applications that automatically extracts, manages, and retrieves relevant long-term context without bloating the model's context window.
Persistent engineering memory & session-continuity control layer for AI coding agents (Claude Code, Codex, Cursor, Gemini), built on the Model Context Protocol. Scans a repo once, remembers architecture/decisions/changes, checkpoints safely, and hands the next agent an exact continuation brief — so agents stop rescanning and losing context.
Lightweight long-term memory retrieval for chat agents using NPMI + Hebbian learning — no vector database or embedding model required.
Relation-aware, calibrated and provenance-preserving context routing for interleaved agent conversations
Temporal context, CRUD tools, and automated deduplication for Mem0 OSS + Qdrant. Fixes time-blindness and conflict resolution gaps.
DBX is an open-source, high-density in-memory database engine built in Go, engineered specifically to serve as the secure memory layer for multi-tenant AI agents. Unlike general-purpose databases that rely on logical access controls, DBX enforces **per-tenant physical isolation** at the kernel level. By leveraging Linux Landlock and `SO_PEERCRED`
Contextra is a high performance AI context engineering platform built in Rust for document ingestion, retrieval, embeddings, memory, prompt orchestration, and multi-LLM application development.
Open-source runtime for LLM agents with persistent memory, relationship history, modeled mood, and evolving identity. Carries state across conversations into prompts for Ollama, OpenAI-compatible backends, or Codex. Inspectable state, with opt-in autonomous research mode.
Official implementation of Learning What to Remember: Resource-Efficient Self-Improvement for Lifelong Agents. TokenWorth learns memory worth from execution feedback.
Claude Code plugin: a knowledge base of your project's conversations and code that builds and updates itself (MemPalace + Graphify)
AI Work Continuity & Cross-Model Agent Handoff System. Captures, verifies, and packages deterministic project state across Claude, GPT-4o, Gemini, Cursor, Antigravity & Local LLMs.
Proven-freshness context store for coding agents - mtime-verified reuse, TF-IDF recall, faithful compression. 100% local, 0 dependencies.
The Experience Lifecycle of lifelong, self-evolving & memory-augmented LLM agents — acquisition, representation, retrieval, consolidation, abstraction, internalization, revision, distribution + governance. Companion to the survey.
[arXiv] ExpHarness: Model-Agnostic Experience Learning through a Trainable Harness
Penpal MCP: a local-first memory layer for AI agents on macOS. Gives Claude Code, Codex and Cursor searchable, cited memory of your browsing, clipboard, windows and past AI sessions, with per-agent grants and redaction. No cloud, no built-in LLM.
Self-hosted memory for LLM agents - one SQLite file, zero services, MIT. Hybrid retrieval (FTS5 + sqlite-vec + RRF) with pain-weighted salience and a knowledge graph. Under matched embeddings, outperforms Mem0 on LoCoMo and LongMemEval. Your data, your provider, your disk.
Memory for AI agents where a wrong fact causes harm. When a medication or an account status changes, Vayl retires the old value, keeps the history, and double-checks risky changes before they apply. Open source, local-first, works with any MCP client or agent framework.
Persistent memory for AI agents in Python. Short-term session memory plus long-term semantic recall, local-first with no services to run, and Redis and Qdrant when you scale. Adapters for LangChain, LangGraph, CrewAI, LlamaIndex and 12 more frameworks.
Install and connect formally verified and executable concepts and rules.
Adaptive memory modality selection for GUI agents: learning when to use text or image memory at each interaction step.
A dependency-free memory kernel for agents: claims with authority, freshness and verifiable provenance. SQLite and the standard library only.
Nightly memory curation for Claude Code on your Mac. Injects repo and client context at session start and keeps it current while you sleep.
Memory for AI agents: beliefs with a source, a certainty and a revision history. SQLite, zero deps, MCP server.
common context MCP — shared, append-only context store for AI agents over MCP: checkpoints, failures, file claims, handoffs and crash recovery. npm: https://www.npmjs.com/package/common-context-mcp
Transparent context reduction for LLMs. A proxy between your agent and your LLM endpoint.
Long agent runs. Under watch.
MemPABench: an interactive benchmark for on-policy evaluation of personal-agent memory for interaction preferences
context handover with increased security, brain child of RELAY_context
The AI that remembers you. Forever. — A living digital entity built in Rust. 68 crates. Self-writing genome. Persistent memory. Constitutional governance. Drop a TOML file, Hydra learns.
Long-term memory for AI coding assistants. Rust CLI with semantic search, Agent Skill integration, and local deployment.
Markdown-first, local-first memory architecture for AI agents with lifecycle, distillation, and human-readable memory artifacts.
Persistent Anticipation Layer for LLM Agents — Pre-compute possible futures during idle time for faster, cheaper and more proactive decisions. Bio-inspired temporal awareness (short/medium/long-term horizons).
Context compiler for agentic AI — token budgets, cache-aware adapters, pytest-style eval layer, JIT references. Framework-agnostic, local-first, deterministic
Architecture documentation for the Firm AI Agent ecosystem — setup-vs-agent-firm + mcp-openclaw-extensions + Memory-os-ai
Context engineering toolkit for LLMs — pack, cache, debug, red-team, and orchestrate context windows. Council of Experts, adversarial testing, immune system, context compiler, drift detection, multi-agent entanglement. TypeScript + Python.
MAGI — Multi-Agent Graph Intelligence. Universal memory server for AI agents. MCP + gRPC + REST API with semantic search, knowledge graph, and behavioral pattern analysis.
Collective AI memory for multi-agent teams. 10-stage GAMR search engine, knowledge graph (Apache AGE), document ingestion, 22+ MCP tools.
Cognitive infrastructure for AI agents
Privacy-first shared-memory governance skill for OpenClaw teams, with safe canonical notes for groups, projects, and dispatcher-led delegation.
MemOcean — A shared knowledge base system for Chinese multi-agent scenarios. Fork of MemPalace.
Hermes Agent 多智能体记忆管理 Skill — 惰性索引、Context注入、记忆回收、主权管理(方案B实现)
A knowledge management platform which acts as a persistent, structured, and token-efficient memory layer for large language models
Stop burning API tokens. 9 algorithmic engines for real-time context optimization, infinite-loop detection, and smart prompt compression.
Event-sourced persistence for AI agents — append-only journal, SHA-256 blob store, knowledge graph, memory API. Part of the Life Agent OS.
⚖️Agentic Context Engineering (https://arxiv.org/abs/2510.04618) implementation using Agent Framework
Cortex System: The Operating System for AI Engineering. Cure Technical Amnesia.
The open-source memory engine behind Cerebro. Persistent memory, learning and reasoning for Claude Code and any MCP client. Local-first.
The self-optimization layer for AI agents.
Local graph-based memory for agents. Memories reinforce when used and fade when ignored.
Agentic Knowledge Base Framework — self-organizing knowledge for AI agents
Errloom: the context engineering framework for shoggoth whisperers; weave beautiful agentic RL tapestries using the Holoware DSL.
MCP server for pūrmemo — AI conversation memory that works everywhere. Save and recall conversations across Claude Desktop, Cursor, and other MCP-compatible platforms.
Personal AI with persistent memory. Turns any LLM into an entity that remembers who you are, learns from your conversations, and grows over time. Local, private, yours.
青龍 Seiryu — AI 記憶與傳承模組。讓 AI 的記憶不再隨視窗關閉而消失。基於 LDRIT(生死遞迴智能理論)設計。
A geometric memory database with importance-stratified shells, HNSW acceleration, and mass-weighted decay scoring. Zero to hero: SQLite → 10M+ records.
CLI-native state persistence for AI agents — tasks, memory, state, and queues. One binary per primitive. No setup. No infrastructure.
Persistent memory for AI agents using git refs. MCP server + Claude Code skill. No vector database, no embeddings, no daemon — branches are scopes, push/fetch is sync.
Experimental pre-1.0 multi-agent memory governance/control plane
🧠 Persistent project memory for AI coding agents. Your agent never forgets, never drifts, never re-asks. Works with Claude, Cursor, Windsurf & more.
TraceCC gives AI agents a CLI-native long-term memory over their own session history. Searchable, scoped, and sized for context windows.
Build the context from anywhere, then bring it to any AI.
Δ Engram is a deterministic reasoning kernel — symbolic AI, a finite state machine, with configurable boundaries and self-improving weights
Rust SDK for MnemeBrain — a belief memory system for AI agents. Supports MnemeBrain Lite and Full APIs.
Persistent local MCP/CLI memory system for AI agents. CPU-only, 98.94–99.15% R@5 on LongMemEval. Named after Mnemosyne (Greek: μνήμη, memory).
Temporal-awareness hook for tool-using LLMs — surfaces freshness verdicts on in-context data so stale information doesn't poison fresh reasoning. Adapters for Claude Code, Antigravity, and any generic harness.
Persistent memory for AI agents. Two lines of code. Works with LangChain, CrewAI, AutoGen, Claude, MCP. Free tier. Self-hostable. MIT.
Polymathera's no-RAG, multi-agent framework for extremely long, dense contexts (1B+ tokens).
Do agents leave footprints? A system for deliberate orientation when minds enter long-lived territories.
Agent context compressor and memory persistent orchestration
One assistant. Many vaults. No bleed. — a free, MIT personal-AI scaffold on Claude Code with governed memory + an IP-clean filter.
🧠 Nexus Mind - Where Intelligence Connects | 三引擎智能记忆系统 | 孕育智慧 · 启迪创新
Persistent memory and replanning for Claude Code. Checkpoints what Claude was doing and injects context at the start of the next session.
A working, tested memory engine for an AI chief of staff: a SQLite fact store with full write history, session commands, enforcement hooks, incident logbook, self-tending rituals, capability registry, mode contracts, and 12-week cycles. Eight staged releases with probes and CI.
Persistent engineering context for AI coding agents. MCP server with auto-hydration on session start.
Fight AI context rot: keep Claude Code (and any LLM) sharp in long sessions with one living Checklist instead of lossy /compact summaries. Persistent working memory + context-window HUD + zero-keystroke auto-clear.
Deterministic state engine for managing conversation state and constraints in LLM applications.
gradio region:us
streamlit region:us
static region:us
A persistent cognitive entity — one agent, one memory, one timeline across all platforms. Self-editing knowledge graph, cross-platform identity, P2P network.
Dynamic Knowledge Graphs with Memory Decay for LLM Agents
Portable semantic memory layer for LLM agents over MCP
Local-first, layered memory kit for AI agents — stdlib-only, zero deps
Multi-agent collaboration engine with persistent, searchable, self-compacting memory. Agents research, critique and adjudicate over a shared SQLite memory with hybrid BM25 + embedding recall. Zero-dependency core, runs locally on Ollama.
Context Transform Engine.
Graph RAG on SQLite for AI agents: vector retrieval + hand-curated wikilink graph + cross-encoder rerank, with a zero-token per-turn memory ledger. Working pilot.
Judgment continuity for AI agents — load your project's scars, taste, and non-negotiables into every session. Judgment-layer companion to edda. (formerly project-doctrine)
The reconciling memory layer for AI agents (MCP server): a new value supersedes the old, "we dropped X" actually retracts X, and history stays queryable — so agents get current facts, not stale ones.