agent-memory README

The project's README from github.com/tigerless-labs/agent-memory. As one markdown file: README.md.

agent-memory: the long-term memory runtime for AI agents

Invariants · Skill · Issues

An agent that closes its session forgets everything it learned in it. agent-memory is the runtime that fixes that, for any agent — not only coding ones. Markdown files in one store are the single source of truth, the SQLite index beside them is a cache you can delete at any time, and Claude Code, Codex CLI, and anything else that can run a shell command share that store.

Retrieval is local and ranked, and it answers with paths rather than pasted text — the agent opens each hit only as deep as the task needs. Writes do not wait for the agent to remember to make them: they fire at conversation boundaries. A sleep-time pass then consolidates and forgets by value, on its own clock. None of it needs an API key.

Two lines, one store

Agent memory has grown along two architectural lines. One builds a retrieval engine — embeddings, a knowledge graph, a ranking pipeline — which finds the right thing, but hands the agent an opaque chunk it cannot inspect and a store it cannot migrate off. The other hands the agent a filesystem — markdown it reads directly, browsable with ls and grep — which is legible and costs nothing to run, but does not rank, and stops scaling the moment the tree outgrows a listing.

agent-memory is the two of them in one store: the retrieval engine indexes a filesystem the agent can also just read. Relations live as links inside the memories, a local index ranks them, and every hit resolves to a whole markdown file on disk. Recall gains the precision of a graph and a vector search without giving up a plain directory an agent can walk — and it stays fast, because nothing in the read path calls a model or crosses a network.

Retrieve by path, then read by level

Recall does not paste text into your context. It answers with an L0 list — one-line abstract, file path, anchor, score — and the agent opens what it wants at the depth the task needs:

mem recall "why files instead of a database"    # L0 list, 8 entries by default
mem read <name> --level outline                 # headings only; or abstract, or full
mem context "why files instead of a database"   # both in one call, top few expanded in full

Index line → abstract → full file → raw material: each rung costs an order of magnitude more than the last, and each is a place to stop. Long files add two free rungs — the anchor that matched, and an outline computed at read time.

Design commitments

The store

$AGENT_MEMORY_STORE/
├── MEMORY.md              root index, one line per memory — the only resident injection
├── config.toml            every tunable; an unknown knob is refused at load
├── schemas/               one file per type: its key fields, the field it groups by, write mode
├── decision/              memories live at <type>/<group>/<name>.md, placed by the schema
│   └── agent-memory/        …/markdown-files-are-the-single-source-of-truth.md
├── archive/               append-only, out of the retrieval surface by default
│   ├── provenance/        distillation evidence, kept forever
│   └── sessions/          full trace copies, in case the host prunes its own
├── dream-reports/         one per sleep: what moved, what was proposed, evidence pointers
├── .index/                fully rebuildable: content-hash manifest, FTS5, access log
└── .state/                runtime state that is not: distillation watermark, write lock

One memory is one file, because the file boundary is the invalidation atom: superseding, weight, and recall all operate on whole files, and a file is either active or invalid with nothing in between. Frontmatter carries the stable name, a one-sentence abstract, the type and its schema fields, status, timestamps, links, weight, and provenance; the body is free markdown.

Proof it works

Measured on LongMemEval-S with a bounded haystack, 120 episodes, claude -p (Haiku 4.5) as host, one calibrated Sonnet 5 judge, two exam replays per arm.

arm pooled accuracy paired vs agent-memory
agent-memory W2 127/240 = 52.9% —
MemCore W2 86/240 = 35.8% +37/−17, p=0.009 · +35/−14, p=0.004
no memory 7/120 = 5.8% +61/−4 · +60/−4, p<0.001

Absolute numbers are not comparable to published LongMemEval scores — the haystack is bounded to 12 sessions per episode, which makes this a write-strategy study rather than a corpus-size one. The system-to-system row differs in write and read together, so it is an end-to-end comparison and licenses no attribution to either half.

One store, three hosts: all 9 ordered writer/reader pairs across Claude Code, Codex CLI, and Hermes pass — what one host's shell writes, another's finds, specifics intact. Pooled net contribution over no memory: 2/36 → 13/36, p=0.0074.

The protocol that decides whether a measurement counts as a result, the full ledger, and the raw run records live in docs/experiments.md and experiments/ in the working tree. They ship with the source, not with git history.

Install

Requires Python 3.12 or higher and uv. There is no release on PyPI yet, so install from a checkout:

git clone https://github.com/tigerless-labs/agent-memory.git
cd agent-memory
uv sync --all-packages

That builds mem, mem-mcp, and mem-hook into .venv/bin. Inside the checkout uv run mem reaches them; put the directory on your PATH so your agents can too — the hook installed in the next section is a bare mem-hook command, and a host that cannot resolve it records nothing:

export PATH="$PWD/.venv/bin:$PATH"

Quick start

mem init

The store defaults to ~/agent-memory-store; export AGENT_MEMORY_STORE only to put it somewhere else, and export it everywhere your agents run, not just in this shell.

Write one memory, find it again, then throw the index away and prove nothing was lost:

mem record --type decision --field project=agent-memory \
  --abstract "Markdown files are the single source of truth" \
  --body "Indexes are rebuildable caches."
mem --json recall "source of truth"
rm -rf ~/agent-memory-store/.index && mem rebuild

Wire it into your agent

mem setup --host claude-code   # or: --host codex

setup probes the host, appends the mem-hook command to its own hook dialect, and leaves the rest of the settings alone — SessionStart injects, Stop and SessionEnd distil, PreCompact evicts. Agents that speak MCP get the same core calls through mem-mcp (memory_recall, memory_read, memory_record, memory_correct, memory_feedback). Anything that can run a shell command needs neither: the CLI is the universal fallback, and it is the wider surface — context, sleep, and the proposal ledger have no MCP tool yet.

Let it sleep

mem sleep --reason host   # consolidate; T0 applies, T1 files a proposal
mem proposals             # what is waiting on you
mem decide <id> --accept

Manage borrows its reasoning from the host CLI you point it at, writes a dream report for the pass, and cannot delete anything unattended.

Develop

uv run pytest -q && uv run ruff check . && uv run mypy

The task lifecycle and the invariants a change must not break are in CLAUDE.md.

License

MIT.

agent-memory (tigerless-labs/agent-memory)

Searching Database…

4.1K ·GitHub
Agent-memory
What is this?agent-memory is a long-term memory runtime for AI agents — it stores memories as plain Markdown files (the source of truth) with a local SQLite/FTS5 index for fast ranked retrieval, so agents like Claude Code and Codex CLI keep what they learned across sessions.
Is this a vector database, or just a folder of markdown files?Neither alone — it's both: markdown files an agent can browse directly with ls and grep, plus a local retrieval engine (BM25/FTS5, with an optional vector plugin fused in) layered on top that ranks them and carries a graph through links, with no separate database to run.
How do I wire this into Claude Code or Codex?A setup command probes your host agent and appends the mem-hook command to its own hook dialect — SessionStart injects memory, Stop/SessionEnd distill new memories, and PreCompact evicts. Agents that speak MCP can instead call mem-mcp's tools (memoryrecall, memoryread, memoryrecord, memorycorrect, memoryfeedback).
Do I need an API key or a separate database to run this?No. You just need Python 3.12+ and uv to install it; there's no LLM client inside the library, so judgement is borrowed from whatever host CLI (Claude Code, Codex, etc.) is already running, and no key or billing surface is needed.
Is there a ready pip-installable package?Not yet — there's no PyPI release, so you install from a git checkout with uv. The CLI is also currently the wider surface: some features like context, sleep, and the proposal ledger have no MCP tool yet.
Does this only work with coding agents?No — it's built for any agent that can run a shell command, not just coding assistants; Claude Code, Codex CLI, and other hosts can all read and write the same store.
Does any of my memory data leave my machine?No — recall runs fully locally and never calls a model or crosses a network; your markdown store lives on disk (default ~/agent-memory-store) and stays greppable, git-able, and portable off the system.
How does the sleep-time Manage layer decide what to forget?A separate consolidation pass runs on its own clock with tiered authority: it can freely add or update memories, but deletion only ever arrives as a proposal you must confirm, so nothing disappears unattended.
Can multiple agents share the same memory store?Yes — Claude Code, Codex CLI, and Hermes were tested writing and reading one shared store, and all 9 ordered writer/reader pairs across the three hosts successfully found each other's memories.
Does it actually improve recall over having no memory at all?On the LongMemEval-S benchmark with claude -p (Haiku 4.5) as host, agent-memory scored 52.9% pooled accuracy versus 35.8% for a competing memory system and just 5.8% with no memory at all.
What happens if I delete the SQLite index?Nothing is lost — markdown files are the source of truth and the index is just a rebuildable cache; running rm -rf .index/ && mem rebuild is enforced by a test to lose zero knowledge.
What are similar projects?okf-agent-memory, ai-memory, qmd, hindsight, leviathan, mcp-memory, EverOS, engram
Which topics is it in?retrieval, bm25, long-term-memory, knowledge-management, privacy-first, agent-memory, knowledge-base, sqlite, local-first, mcp-server, markdown, codex
Creator
tigerless-labs
Size
117.5 MB
Stars
4.1K
Updated
CommitOct 16
Age
1mo
Commits · 90 Days