If you already work with an AI coding agent, such as Claude Code, Cursor, Codex or GitHub Copilot, it can operate dPanel for you through dnocs, the dPanel command-line tool. The agent talks to the dPanel assistant with dnocs ai chat, and every change still goes through an approval with dnocs ai approve. It can also read this documentation from llms.txt and llms-full.txt.
How it works
- You install dnocs and sign in yourself. See Use AI in the terminal.
- You tell your agent how to use dPanel (the instructions below).
- The agent asks the dPanel assistant for what you want, for example "Deploy my repository to server 145 and put it on shop.example.com".
- The dPanel assistant proposes each change with an action ID (
aipa_…). - The change runs only when
dnocs ai approve <action-id>is run.
The agent never calls the dPanel API or your cloud provider directly, and never changes a server by hand. Everything goes through the dPanel assistant, so every change is proposed, approved and recorded the same way as in the web app.
Give your agent the instructions
Paste this into your agent's instructions file, such as AGENTS.md, CLAUDE.md or the Cursor rules, or into the chat:
## dPanel
dPanel (https://dnocs.devetek.app) runs my servers, applications, databases, domains and load balancers.
Operate it only through the dnocs CLI and its assistant:
- `dnocs ai chat "<request>"` asks the dPanel assistant. Give it IDs and names. It reads what it needs
and, for any change, proposes it and prints an action id (aipa\_…). Nothing changes yet.
- `dnocs ai chat --conversation <aic_id> "<request>"` continues the same conversation.
- `dnocs ai approve <aipa_id>` runs one proposed change, exactly as proposed. `--reject` declines it.
Show me each proposal and wait for my explicit yes before approving.
- Approve one change at a time. Follow its job with `dnocs execution list --server-id <id> --limit 3`
until it says succeeded or failed; if it failed, ask the assistant why.
- Never call the dPanel API or a cloud provider directly, never change a server by hand, and never put a
password, token or key in a chat message. Secrets I must supply are prompted by `dnocs ai approve`;
generated secrets are printed once by `dnocs ai approve`: hand them to me immediately.
- Do not choose ports: dPanel picks a free one. For an app without a .devetek definition, tell the
assistant the language and version, build commands and start command.
- How-to questions: read https://dnocs.devetek.app/llms.txt and the pages it links to.
The dPanel team uses a fuller version of these instructions, the dpanel-cli-ai skill, to test dPanel end to end with AI agents (a server, MariaDB, a WordPress deploy, an HTTPS load balancer and the clean-up, all through dnocs ai). The rules above are its core.
Good requests to give your agent
- "Using dPanel, deploy this repository to server 145 and give it the domain shop.example.com with HTTPS."
- "Using dPanel, find out why the last deploy of application Shop failed and propose a fix."
- "Using dPanel, create a PostgreSQL database for this app on server 145 and set the app's database variables."
- "Using dPanel, open port 8080 on server 145 for the staging API."
More requests in the prompt cookbook.
Docs for AI: llms.txt and llms-full.txt
The dPanel documentation is published in two plain-text files that AI tools can read:
| File | What it holds |
|---|---|
| llms.txt | A short index: every documentation page with its title, link and one-line description |
| llms-full.txt | The whole English documentation as Markdown, page after page |
Point your agent or chatbot at llms.txt to find the right page, or give it llms-full.txt to answer from everything at once. The dPanel assistant itself answers how-to questions from llms-full.txt.
Frequently asked questions
Does my agent need its own dPanel account?
No. It uses the dnocs session you signed in with, so it works with your access and your role. Anything it approves runs as you.
Can my agent skip the approval step?
Not in dPanel: every change still needs dnocs ai approve. But an agent that is allowed to run commands can run that command itself, so tell it to wait for your yes, as in the instructions above.
Can I give the agent my browser session instead of signing in?
Sign in with dnocs auth login yourself instead. A session is as powerful as your password; keep it out of chats, files you share, and screenshots.
Which agents work?
Any agent that can run terminal commands, such as Claude Code, Cursor, Codex, Gemini CLI or GitHub Copilot in agent mode.