Docs for AI tools
These docs are published so that language models, coding assistants (Claude, ChatGPT, Cursor, Copilot) and autonomous agents can read them completely and accurately.
Files
| File | Contents |
|---|---|
/en/llms.txt | Every page with a short description and a link to its Markdown version, in the llmstxt.org format |
/en/llms-full.txt | The whole documentation in one text file, in sidebar order |
/en/<page>.md | The Markdown version of any page, e.g. /en/cloud-api.md |
/llms.txt and /llms-full.txt | The same files in Persian, the site's default language |
| OpenAPI documents | The machine-readable description of each API; listed in the API reference |
Both languages say the same thing; English tends to work best with most models.
The "Copy page" button
At the top of every page, Copy page copies the page's Markdown so you can paste it into a chat. The menu next to it opens the page directly in Claude or ChatGPT, or shows the Markdown version.
Example prompt
Read https://docs.ariaservice.net/en/llms.txt and the pages it links that you need.
Using the Arianet Cloud API, write a Python script that creates a server named web-1
in datacenter ovh with plan 3 and image ubuntu-24.04, waits for the create action
to finish, and prints its public IPv4. Read the token from ARIANET_TOKEN.
Rules for autonomous agents
If you build an agent that works on Arianet with your token:
- Least privilege: give the agent its own token with the fewest scopes it needs; only
readscopes for read-only work. - Keep the token away from the model: hold it in the tool's environment, not in the prompt or chat history.
- A human approves costly or irreversible actions: creating, deleting, reinstalling, renewing and sending SMS cost money or erase data; ask before running them.
- Retry safely: send an idempotency key with orders, and read the state before retrying after a
504. - Wait for the end:
201/202means the work started, not that it finished; follow the action. - Respect rate limits: honour the
RateLimit-*andRetry-Afterheaders. - Source of truth: rely on the API's OpenAPI document for exact field names, not on the model's guess.