# Agent NAS > Agent NAS is network-attached storage designed around the run: every agent run works in an isolated branch and reaches data through a short-lived, narrow lease; every change can be traced and undone; and memory is plain files with provenance. It speaks what agents already speak: files, Git, S3 and MCP. Status: concept and draft specification v0.2, November 2026. No reference implementation yet. Every page is also published as Markdown: append `index.md` to a page URL or send `Accept: text/markdown`. The links below point to the Markdown versions. Content licensed under CC BY 4.0. ## Start here - [What is Agent NAS?](https://agentnas.com/what-is-agent-nas/index.md): Agent NAS is network-attached storage built around the agent run. Its definition, the six conditions for being one, what it is not, and who does not need it. - [The nine concepts](https://agentnas.com/concepts/index.md): Agent NAS proposes nine named concepts in three clusters: Work, Authority and Knowledge, each with a mechanism, a way to start small and honest limits. - [Architecture](https://agentnas.com/architecture/index.md): The Agent NAS architecture: four layers plus optional Sync, built on six nouns. The Store contract, the journey of a run and the deployment forms. - [Interfaces](https://agentnas.com/interfaces/index.md): Agent NAS serves data over mounts, Git and S3, and control over MCP, HTTP and a CLI. Nine MCP tools face agents, and every HTTP call is POST with JSON. - [Specification](https://agentnas.com/spec/index.md): Agent NAS specification draft v0.2: its components, five conformance levels from L0 to L4 with executable acceptance tests, and the conformance.json format. ## The nine concepts - [Agent Home](https://agentnas.com/concepts/agent-home/index.md): A directory per role, with history. Plain files you can take with you. - [Agent Branch](https://agentnas.com/concepts/agent-branch/index.md): Each run works on a copy-on-write branch. Only a reviewed merge reaches the main line. - [Flight Recorder](https://agentnas.com/concepts/flight-recorder/index.md): A per-run, hash-linked record that includes what the run read. - [Capability Lease](https://agentnas.com/concepts/capability-lease/index.md): A narrow authorization for one run, with an expiry and budgets. Child leases can only be narrower. - [Human Gate](https://agentnas.com/concepts/human-gate/index.md): Appears only before irreversible, outward-facing or privilege-raising actions, and shows the real diff. - [Trust Labels](https://agentnas.com/concepts/trust-labels/index.md): Two small labels plus session taint enforce the Rule of Two. - [Agent Memory](https://agentnas.com/concepts/agent-memory/index.md): Notes you can read and edit are the truth. Indexes are a cache you can throw away. - [Budgeted Reads](https://agentnas.com/concepts/budgeted-reads/index.md): Read a file as an outline, a section, a line range or in full, degrading to fit the budget. - [Cascade Forget](https://agentnas.com/concepts/cascade-forget/index.md): Delete derivatives along lineage, handle history by retention class, and report what it could not do. ## Reference - [Security](https://agentnas.com/security/index.md): The Agent NAS security model: what it protects and does not, the Rule of Two at the storage boundary, ten threats with controls, and honest limits. - [Use cases](https://agentnas.com/use-cases/index.md): Agent NAS use cases: overnight agent runs, switching frameworks without losing memory, shared team memory, incident review, forgetting, and when to skip it. - [Comparison](https://agentnas.com/compare/index.md): How Agent NAS relates to NAS vendor agents, agent file systems, forkable storage, memory tools and sandboxes: what it borrows, and the five things that are new. - [Roadmap](https://agentnas.com/roadmap/index.md): The Agent NAS roadmap: concept and draft spec now, reference implementations of Lite and Overlay next, then later work, and what we will not do. - [Design principles](https://agentnas.com/principles/index.md): The eight design theses of Agent NAS and its seven bets: the evidence for each, how the design degrades if a bet is wrong, and what should stay stable. - [FAQ](https://agentnas.com/faq/index.md): Common questions about Agent NAS: what it is, how it differs from NAS vendors’ agents, prompt injection, forgetting, when not to use it, and project status. - [Glossary](https://agentnas.com/glossary/index.md): The Agent NAS glossary: the nine concepts and supporting terms such as principal, run, read manifest, session taint, retention class and crypto-shredding. - [About](https://agentnas.com/about/index.md): About the Agent NAS project: status, license, governance, how to cite and take part, and the machine-readable resources for agents (llms.txt, MCP, OpenAPI). ## Machine interfaces - [MCP](https://agentnas.com/mcp/server-card): Read-only MCP server card (Streamable HTTP endpoint: https://agentnas.com/mcp). Tools: search_docs, get_page, list_pages, get_briefing. - [OpenAPI](https://agentnas.com/openapi.json): OpenAPI description of POST /api/search (JSON in, JSON out). - [llms-full.txt](https://agentnas.com/llms-full.txt): The complete documentation in one Markdown file. - [sitemap.xml](https://agentnas.com/sitemap.xml): All pages in every language, with hreflang alternates. ## Optional - [简体中文](https://agentnas.com/zh-cn/llms.txt) - [繁體中文](https://agentnas.com/zh-tw/llms.txt) - [日本語](https://agentnas.com/ja/llms.txt) - [한국어](https://agentnas.com/ko/llms.txt) - [Español](https://agentnas.com/es/llms.txt) - [Français](https://agentnas.com/fr/llms.txt) - [Deutsch](https://agentnas.com/de/llms.txt) - [GitHub](https://github.com/xwzy/agentnas.com)