Repolode
Submit
tigerless-labs

tigerless-labs/agent-memory

Long-term memory runtime for AI agents that stores memories as plain Markdown, ranks recall locally, and shares one store across Claude Code, Codex CLI and other shell-capable agents.

View on GitHub
Stars
4,310
Forks
204
Open issues
9
Last commit
11 hours ago

What is it?

agent-memory gives AI agents memory that survives after a session ends. Memories are stored as Markdown files in one directory, and the SQLite index beside them is a rebuildable cache, so deleting the index loses no knowledge. Claude Code, Codex CLI, Muse Code and any tool that can run a shell command read and write the same store.

Recall is local and ranked. It returns a list of one-line abstracts with file paths, and the agent opens each hit only as deep as the task needs, from outline to full file to the raw cited session messages. Writes are triggered at conversation boundaries such as session stop and compaction, and a separate sleep-time pass consolidates memories and proposes deletions for user confirmation. Superseded memories are kept, and recall can answer as of a given date.

The library itself contains no LLM client; judgement is delegated to the host agent's own CLI. Optional vector search via FastEmbed and CJK tokenization via jieba are available as extras.

Key features

  • 01Markdown files are the source of truth; the SQLite index can be deleted and rebuilt.
  • 02Local BM25 recall with optional vector fusion via reciprocal-rank fusion.
  • 03Layered reading from abstract to outline, full file and raw cited messages.
  • 04Automatic writes at session boundaries, with sleep-time consolidation that requires confirmation for deletions.
  • 05Shared store across Claude Code, Codex CLI and Muse Code, plus an MCP server option.
  • 06Point-in-time recall with superseded memories kept in the chain.

Use cases

  • Giving a coding agent persistent project decisions across separate sessions.
  • Sharing one memory store between several agent hosts on the same machine.
  • Keeping agent memory in greppable, git-versionable files that stay portable off the tool.
  • Adding Chinese-language recall to an agent memory with the optional CJK tokenizer.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/tigerless-labs/agent-memory.git
  2. 2

    Install dependencies from the checkout with uv

    uv sync --all-packages
  3. 3

    Initialize the memory store

    mem init
  4. 4

    Wire it into Claude Code

    mem setup --host claude-code

Things to know

  • There is no PyPI release yet, so installation requires a source checkout.
  • Requires Python 3.12 or higher and uv.
  • Version 0.1.0 is early, and Codex CLI setup may need extra sandbox or trust approval.

In collections

Long-term memory for AI agentsClaude Code tools, skills & pluginsOpen-source AI agent frameworks & appsAI coding assistants & agentsMCP serversPrivacy & local-first toolsDatabases & SQL toolsStatic site generators & documentation tools

Topics

#agent-memory#ai-agents#long-term-memory#markdown#local-first#retrieval#sqlite#mcp#claude-code#codex

FAQ

What is agent-memory?
Long-term memory runtime for AI agents that stores memories as plain Markdown, ranks recall locally, and shares one store across Claude Code, Codex CLI and other shell-capable agents. agent-memory gives AI agents memory that survives after a session ends. Memories are stored as Markdown files in one directory, and the SQLite index beside them is a rebuildable cache, so deleting the index loses no knowledge. Claude Code, Codex CLI, Muse Code and any tool that can run a shell command read and write the same store.
Is agent-memory free to use? What license does it have?
agent-memory is open source under the MIT license. Check the license terms before using it in commercial products.
Is agent-memory still maintained?
Actively maintained. The last commit to agent-memory was on Nov 6, 2026.
What can I use agent-memory for?
Giving a coding agent persistent project decisions across separate sessions. Sharing one memory store between several agent hosts on the same machine. Keeping agent memory in greppable, git-versionable files that stay portable off the tool. Adding Chinese-language recall to an agent memory with the optional CJK tokenizer.
How do I install agent-memory?
Clone the repository: git clone https://github.com/tigerless-labs/agent-memory.git Install dependencies from the checkout with uv: uv sync --all-packages Initialize the memory store: mem init Wire it into Claude Code: mem setup --host claude-code
Where can I get agent-memory?
The source code and installation instructions are on GitHub: https://github.com/tigerless-labs/agent-memory

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects