Intertrace Docs
Sign in
Browse the docs

Set up with AI

Let your coding agent put your AI agent behind Intertrace. Copy one prompt, paste it into Claude Code, Cursor, Codex or Copilot, and review the change.

Paste into your agent

Copy and run

Paste into your agent

  • or any coding agent
Set up Intertrace in this repo: read https://docs.intertrace.ai/agent-setup.txt and follow its steps exactly.
My runtime key is in the INTERTRACE_RUNTIME_API_KEY environment variable. Never write it into a file, a commit or this chat.

Your agent reads the full steps from the link. If it can't open links, use Full prompt: the same steps, inline. Plain text: /agent-setup.txt

Your agent will

  1. 01Find every place your app creates a model client
  2. 02Point each one at Intertrace, reading the key from your environment
  3. 03Handle Intertrace's answers: a block, an approval, a budget
  4. 04Check tool calls before they run, if your agent runs its own tools
  5. 05Send one normal request and one test attack to prove it works
  6. 06Add short rules to your AGENTS.md so it stays that way

How it works

  1. 1

    Create your agent and its key

    In the dashboard, open Connect → An agent you built. It gives you a runtime key, and the same prompt with your agent's details already filled in.
  2. 2

    Put the key in your environment

    Set INTERTRACE_RUNTIME_API_KEY in your shell or your project's .env. The prompt never contains the key, so there's nothing to leak if you share it.
  3. 3

    Paste the prompt into your coding agent

    Open the project in Claude Code, Cursor, Codex or Copilot and paste the prompt above. It finds your model clients and makes the change.
  4. 4

    Review and run

    When it's done it tells you which files it changed, how to run your agent and what its check showed. Review the diff like any other change, then commit it.

Safe by design

  • Never writes your key into code, config, logs or commits
  • Never sends your model traffic around Intertrace
  • Doesn't change your Intertrace settings or policies
  • Asks before touching CI, deployment or secret stores
  • Keeps your agent's own tools exactly as they are
  • Stops and asks you when the code doesn't match what it expects

Keep it in place

The prompt adds a short Intertrace section to your repository's AGENTS.md (or CLAUDE.md, and Cursor's rules if you use Cursor). Coding agents read that file, so future edits keep your model clients pointed at Intertrace and handle its answers the same way. You can copy the rules on their own from the Rules for your repo tab above.

Good to know

Python, Go and other languages work too: the model client change is the same, and tool checks use the HTTP API. TypeScript projects can use @intertrace-runtime/runtime for tool checks.

Prefer to do it by hand? Follow the Quickstart or the Agent frameworks guide. The prompt does the same steps.

For tools that read docs themselves, the prompt is plain text at /agent-setup.txt and listed in /llms.txt.