Duin Governed Agent Memory

DUIN

**English** · [简体中文](README.zh-CN.md) · [日本語](README.ja.md)

Agents that earn your trust.

**Every agent you run writes its own memory, grades its own judgment and asks for its own permissions. DUIN is a harness that makes it earn all three, in files you own.**

DUIN is an open harness for personal agents: long-term memory, a working model of your world, and the rules for what an agent may do, kept as Markdown in a folder you own and governed at every step. Memory and autonomy are both earned: a fact is labelled at capture as yours or as inferred and proven over sessions before it becomes a rule, and the agent asks before it acts outside your notes. Local, MIT, no account.

[Download](https://github.com/Mentis-lab/duin-governed-agent-memory/releases/latest) · [The harness](#the-harness) · [With Claude Code](#with-the-agents-you-already-run) · [Getting started](docs/getting-started.md) · [Architecture](docs/architecture.md) · [FAQ](docs/faq.md) · [Discussions](https://github.com/Mentis-lab/duin-governed-agent-memory/discussions) · [Security](SECURITY.md)

[](https://github.com/Mentis-lab/duin-governed-agent-memory/actions/workflows/ci.yml) [](https://github.com/Mentis-lab/duin-governed-agent-memory/releases/latest) [](LICENSE)

The author's own vault, about 1,200 notes, recorded live: zoom the brain map in and out, filter the Explorer to a DUIN note, open it, and ask in its context — the composer says asking in context and the answer is grounded in that note. Answered by DeepSeek V4 Flash; map la