---
url: /ai.md
description: >-
  How AI assistants and agents read these docs: llms.txt, llms-full.txt and a
  Markdown copy of every page.
---

# AI-readable docs

These docs are written to be read by AI as well as by people. Everything is Markdown.

## Three ways to read them

| What | Address | Use it for |
| --- | --- | --- |
| **Index** | [`/llms.txt`](/llms.txt) | A list of every page with a one-line description. Give this to an agent that can fetch pages as it needs them. |
| **Whole site** | [`/llms-full.txt`](/llms-full.txt) | Every page in one file. Give this to a chat as its source. |
| **One page** | Add `.md` to any page address | The Markdown of that page alone, for example `/operate/role-ai.md` |

## Rules for an AI that reads these docs

1. **Respect the status labels.** A feature marked Planned is not available. A feature marked In review is written and not released. Never describe either as live.
2. **Do not invent.** If these docs do not answer a question, say so.
3. **Use the product's words.** See [Words we use](../start/words-we-use.md).
4. **These docs describe the product.** They hold no client data. A client's own map is read through MCP or the CLI, with that person's access.

## Good uses

* **Train a new advisor.** See [AI trainer](../training/ai-trainer.md).
* **Answer "how do I" questions** while you work in the app.
* **Give a coding agent context** before it works on DutyGraph.
* **Draft client-facing explanations** that match how the product really works.

## Where the pages come from

| Section | Source in the repository |
| --- | --- |
| Start, Operate, Playbooks, Offers, Build, Training | `developer-docs/` |
| Product handbook | `gitbook/product/` |
| Advisor guide | `docs/guide/` (the same pages the app shows under Help) |
| CLI commands | `docs/cli/README.md` |
| MCP tools | The tool definitions in `server/`, read when the site is built |

One page has one source. To change a page, change its source.
