Skip to content
☰
Documentation navigation +
implemented

Use the AI intake prompt with your own assistant

For private or complex projects, the preferred path is to let your own AI assistant (Claude, ChatGPT, Gemini, a coding agent) read your material and write a compact IntakeSpec. Only that JSON reaches Alive.

Your assistant's provider still processes whatever you give it, under that provider's terms. Decide what you share accordingly.

  1. Get the prompt from the app (Use my AI assistant), from GET /v1/prompts/intake, or with the CLI:

    uv run alive prompt intake | head -5

  2. Paste the prompt into your assistant together with only the material you authorize.

  3. Save the JSON it returns as intakespec.json and check it locally (free, no account):

    uv run alive intake validate intakespec.json

    Validation catches what assistants commonly get wrong: claims citing sources that were never listed, "observed" claims without a processed source, or production_access_approved set to anything but false.

  4. Import it in the app or with POST /v1/projects/{id}/intakes, review, and confirm.

A prompt never guarantees valid output; schema validation in Alive does. The schema is published at https://www.alivehealthtech.com/schemas/intakespec/1.0.0.json and printed by uv run alive schema intakespec.

For projects that already have a plan recommending simulation, uv run alive prompt worldspec prints the advanced prompt for drafting a WorldSpec from the model catalog.