Find a memory tool that works with your bots

Persistent memory for agent pipelines and chat clients: REST API, knowledge graph and autonomous consolidation, with remote MCP over OAuth.

Memory and context engine with an app, API and MCP; hosted, with a documented fully-local mode.

Local persistent memory store with MCP, HTTP API and session import from many coding agents.

Agent memory that learns: self-hosted server with per-bank MCP endpoints and native coding-agent installers.

Memory library and service for stateful agents, with a hosted MCP at mcp.honcho.dev and self-hosting.

Notes-as-memory: your AI reads and writes plain Markdown files you can open in Obsidian, indexed locally.

Hosted shared memory across ChatGPT, Claude, Cursor and 20+ AI tools via browser extension, apps and a remote MCP server; paid, with a 7-day trial.

Captures what your coding agent does, compresses it and injects relevant context into future sessions.

Personal memory across agents, with CLIs (including a Codex CLI) and hosted or local storage.

Hosted context layer for agents (Zep Cloud). The GitHub repo holds examples and integrations, not the service.

Official Qdrant MCP server used as a semantic memory layer: store and find information in a Qdrant collection.

Centralized, remote-accessible version of the Cline "memory bank" file pattern, served over MCP with per-project isolation.

MCP server for the hosted Graphlit platform: ingest content from many sources into a searchable project.

Desktop memory and context manager with MCP integrations, a browser extension that imports chats, and thread import from coding tools.

Open-source local AI memory system with pluggable retrieval (ChromaDB by default) and auto-save hooks for coding agents.

Context database for agents from Volcengine that unifies agent memory, knowledge retrieval and skills.

MCP server for Cognee memory; by default each client gets its own dataset so agents do not overwrite each other.

Local-first, Markdown-native memory layer for agents, with a cloud option and an MCP package (everos-mcp).

Research memory operating system for personalized agents (EMNLP 2025), with an MCP server.

Memory-bank MCP server that builds a project-specific knowledge graph (decisions, progress, patterns) in a per-workspace SQLite DB.

Read, write, search and edit Obsidian vault notes, tags and frontmatter over MCP via the Local REST API plugin.

MCP server exposing Graphiti knowledge-graph memory, with HTTP endpoint at /mcp/ and Docker Compose setups.

LLM-agnostic memory layer turning agent execution into structured persistent state; hosted MCP for coding agents.

Unified memory for coding agents from Zilliz: Markdown files are the source of truth, indexed in Milvus, with plugins per agent.

Knowledge-graph memory with semantic retrieval and temporal awareness, using Neo4j for both graph and vector search.

Knowledge-graph memory (entities, relations, observations) stored in a local JSONL file. The reference implementation many forks build on.

Open-source AI memory platform: builds knowledge graphs plus vectors from your data for agents.

Memory management kit for agents from the AgentScope team, with plugins for coding agents.

Universal memory layer for agents with a native MCP server and Python client.

Self-hostable "personal AI OS" with memory shared across coding agents and 50+ app integrations via MCP.

Fork of the official memory server focused on local development: persistent knowledge graph in local files.

Template MCP server for long-term agent memory using the Mem0 library and Postgres.

MCP server for the mem-agent model: an Obsidian-like Markdown memory managed by a small local LLM, with importers for ChatGPT and other exports.

MCP server for the Milvus vector database: search, query and manage collections as a semantic memory store.

Self-deployed remote MCP memory server on Cloudflare Workers, Vectorize, D1 and Durable Objects.

Persistent context management for coding assistants: saves session context, checkpoints and decisions to SQLite.

Persistent knowledge-graph memory for MCP backed by libSQL (local SQLite file or remote Turso).

MCP server that reads and writes an Obsidian vault through the Local REST API community plugin.

Letta background agent that watches Claude Code sessions, builds memory over time and feeds it back.

Knowledge-graph memory MCP server backed by Neo4j (local or AuraDB), from Neo4j Labs.

Official MCP server giving clients collection and document tools on a Chroma vector database (ephemeral, persistent or cloud).

MCP server to retrieve context from a hosted Pinecone Assistant.

Widely used memory layer for AI agents and apps: Python/TS SDKs, hosted platform and a self-hosted server.

Framework for real-time, temporally-aware knowledge graphs for agents; the engine behind Zep.

Stateful agents with self-editing memory (formerly MemGPT). Active source has moved to letta-code.

Single-file memory layer for agents: portable, versioned memory without a database server.

All-in-one embeddings database and AI framework with web and MCP APIs; usable as a local semantic memory store.

Memory OS for LLMs and agents with hybrid retrieval and skill reuse; cloud plugin and local plugin.

Letta's current agent harness: terminal UI, App Server and runtime for agents with persistent memory.

Personal assistant that builds multi-type memory from screen observation and conversation; server plus client SDK.

User-profile-based long-term memory for chatbot apps, with an MCP server in src/mcp.

Open-source memory layer for autonomous agents built on knowledge graphs.

LangChain library for agents to extract, store and update long-term memory, built on LangGraph stores.

Research library (ICLR 2026) for lightweight memory-augmented generation, with an optional MCP server.

Research library for agentic memory (Zettelkasten-style linked notes) for LLM agents.

Evidence-first local memory runtime for agents: admission policies, temporal versions and citations, with an MCP server extra.

Redis's managed memory layer for agents (part of Redis Iris), with a benchmark harness in the repo.

Predictive memory layer for agents on MongoDB, Qdrant and Neo4j with caching, custom schemas and GraphQL.