# memfor

> Deterministic, local memory for AI agents.

memfor is the home of agentware, an open-source (Apache-2.0) memory and work loop for AI coding agents such as Claude Code and OpenAI Codex. Knowledge is plain Markdown and JSON in a folder the user owns; a validating toolkit is the only writer of the index; recall is BM25 from Python's standard library. The memory toolkit never calls an LLM. A hosted team tier (mem for Teams) is planned and not built.

## Start
- [Quickstart](https://memfor.ai/docs#quickstart): clone, pin a commit, run your agent
- [Agent setup prompt](https://memfor.ai/docs#agent-setup): text to give your coding agent; it asks before installing anything
- [Source](https://github.com/r5rana/agentware): the agentware repository

## Proof
- [Benchmark ledger](https://memfor.ai/benchmarks): LongMemEval-S Recall@5 0.9140 (fractional recall, 470 answerable questions, commit 42d58a0), reproduction steps
- [Comparison](https://memfor.ai/compare): capability comparisons, dated and sourced; others' benchmark headlines labelled by metric and never ranked

## Products
- [mem for AI](https://memfor.ai/ai): the engine, as a datasheet
- [mem for Code](https://memfor.ai/code): memory for Claude Code and Codex
- [mem for Teams](https://memfor.ai/teams): early access, not built
- [mem for Learning](https://memfor.ai/learning): exploring, no product yet

## Optional
- [Security](https://memfor.ai/security)
- [Changelog](https://memfor.ai/changelog)
- [Privacy](https://memfor.ai/privacy)