LLM-native docs (llms.txt)

AI-assisted integration

This guide is published in forms an AI coding assistant can consume directly, so "build me the AIOTIC receive endpoint" produces code that matches the real contracts instead of a plausible guess.

File	Content	Use
/llms.txt	Index: one line per page with a summary, plus links to the spec and samples	Paste the URL into your assistant; it fetches what it needs
/llms-full.txt	The whole guide as one Markdown file, in reading order	Add to a project's context / knowledge files
/openapi.yaml	The public OpenAPI 3.1 document, including the outbound webhooks	Code generation, request validation, schema-aware assistants
/samples/*.json	Real payload samples (ERP receive request, order status)	Fixtures for tests
<page URL>.md	Every page is also served as raw Markdown at …/path.md	Selective context

How to use it

Claude Code / Cursor / Copilot chat

Read https://developers.aiotic.ai/llms.txt and then implement an ASP.NET minimal-API endpoint
that receives AIOTIC orders according to the "ERP receive endpoint" contract, including
X-API-KEY verification and idempotency on request_id.

Project knowledge

Add https://developers.aiotic.ai/llms-full.txt to the project's documentation sources (or
download it into docs/vendor/aiotic.md) so every prompt in this repo has the contracts available.

CLAUDE.md / AGENTS.md

# In the repository of your integration service:
echo "AIOTIC integration contracts: https://developers.aiotic.ai/llms-full.txt (OpenAPI: /openapi.yaml)" >> CLAUDE.md

For richer interaction — searching the guide, pulling one endpoint's schema, fetching examples, or (opt-in) inspecting your own test tenant — use the MCP server.

What the assistant should know

The most common mistakes we see in generated integrations, all covered in the docs the assistant will read:

• treating the HTTP status of the receive endpoint as the outcome (only success counts),
• forgetting idempotency on request_id,
• re-uploading the entire catalog on a schedule instead of syncing changes,
• assuming a corrections/cancel endpoint exists (it does not yet — see Headless limits),
• polling a single order in a tight loop instead of with backoff.

Freshness

llms.txt and llms-full.txt are regenerated on every build of this site from the same Markdown sources, and the OpenAPI document is regenerated from the backend's spec. The footer of every page states the API version the content was verified against.