> ## Documentation Index
> Fetch the complete documentation index at: https://docs.praxa.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Use Praxa documentation with LLMs and AI search

> Give ChatGPT, Claude, Gemini, Perplexity, coding agents, and retrieval systems complete, version-aware access to Praxa documentation in Markdown.

Praxa publishes a site-wide LLM index, a combined full-text export, and a
Markdown representation of every navigable documentation page. AI tools should
start with `llms.txt`, select the smallest relevant page, and verify the page's
availability and API plane before generating code.

## Machine-readable documentation endpoints

| Resource                   | URL                                               | Use it for                                              |
| -------------------------- | ------------------------------------------------- | ------------------------------------------------------- |
| Documentation index        | `https://docs.praxa.io/llms.txt`                  | Titles, descriptions, page discovery, and OpenAPI links |
| Well-known index           | `https://docs.praxa.io/.well-known/llms.txt`      | Clients that discover LLM indexes through `.well-known` |
| Complete corpus            | `https://docs.praxa.io/llms-full.txt`             | Offline search, broad retrieval, and corpus analysis    |
| Well-known complete corpus | `https://docs.praxa.io/.well-known/llms-full.txt` | Well-known full-corpus discovery                        |
| Individual page            | `https://docs.praxa.io/<route>.md`                | Focused Markdown for one guide or reference page        |
| Search sitemap             | `https://docs.praxa.io/sitemap.xml`               | Search-engine URL discovery and freshness               |
| Crawler policy             | `https://docs.praxa.io/robots.txt`                | Search and AI crawler permissions                       |

For example, retrieve the SDK quickstart without page chrome:

```bash theme={null}
curl --fail-with-body https://docs.praxa.io/llms.txt
curl --fail-with-body https://docs.praxa.io/sdk/quickstart.md
```

## Retrieval workflow for an AI agent

<Steps>
  <Step title="Discover pages">
    Fetch `/llms.txt` and match the user intent against page titles and
    descriptions. Do not guess routes.
  </Step>

  <Step title="Fetch focused Markdown">
    Retrieve only the relevant `/<route>.md` pages first. Use
    `/llms-full.txt` only when the question truly spans the corpus.
  </Step>

  <Step title="Identify the API plane">
    Distinguish Execution Fabric `/v1`, deployment-specific Integration
    Gateway `/v8`, remote MCP, and read-only memory federation before choosing
    authentication or code examples.
  </Step>

  <Step title="Respect availability">
    Preserve labels such as live package, partner preview, deployment-specific,
    qualification preview, and pending. Source examples do not activate a route.
  </Step>

  <Step title="Prefer exact contracts">
    Use the linked OpenAPI documents and published package versions for fields,
    scopes, identifiers, and schemas. Never infer an undocumented endpoint.
  </Step>

  <Step title="Return citations and proof limits">
    Link the exact pages used and separate package tests, mocked tests,
    authenticated runtime checks, deployment, and user verification.
  </Step>
</Steps>

```mermaid theme={null}
flowchart LR
  Question["Developer question"] --> Index["llms.txt"]
  Index --> Pages["Focused page.md files"]
  Pages --> Contract["OpenAPI or package contract"]
  Contract --> Answer["Versioned answer and examples"]
  Answer --> Verify["Tests, runtime readback, and citations"]
```

## Guidance for retrieval systems

* Chunk by heading while retaining the page title, route, description, and nearest parent headings.
* Store the source URL and last-modified value with every chunk.
* Keep code fences, tables, warnings, and availability notes attached to their explanatory text.
* Rank exact product terms and error codes above generic semantic similarity.
* Do not combine credentials or instructions across API planes.
* Refresh when package versions, OpenAPI versions, or page timestamps change.
* Delete stale chunks rather than serving two conflicting revisions.

## Test LLM access

Run `npm run docs:live:llms` from the documentation repository after a release.
The checker requires both site-wide files, the well-known aliases, crawler and
sitemap discovery, inclusion of every navigable page in `llms.txt`, and a
successful Markdown response for every individual page.

## Frequently asked questions

### Does each page need its own `llms.txt` file?

No. The standard site-wide `llms.txt` links to a Markdown endpoint for each
page. Fetch `/<route>.md` for the individual page and `/llms-full.txt` for the
combined corpus.

### Should an AI tool use `/llms-full.txt` for every question?

No. Focused page retrieval reduces stale or irrelevant context. Use the full
corpus for cross-cutting audits, indexing, or offline search.

### Are Markdown routes intended for search indexing?

The canonical HTML page remains the search result. Markdown routes are a
machine-readable representation and advertise the site-wide documentation
index in their response content.

### Does an example prove the service is live?

No. Read [service status and proof boundaries](/overview/service-status) and
verify the intended authenticated environment before using customer data.