Jinn

Jinn runs agent work as a call. You send a prompt and a folder. An agent works on it in its own Linux machine. You get back the files you asked for.

You define the work once, as a function. Then you run it as often as you like, from the CLI, the API, the SDKs or the console.

Quickstart

1. Sign in at app.usejinn.com. Jinn is invitation-only: an account owner invites your email.
2. Add a provider: your model vendor's API key and the model to use. Open Providers in the console.
3. Install the CLI and sign it in:

curl -fsSL https://docs.usejinn.com/install.sh | sh
jinn login

1. Write function.json. Put your provider's id in provider (the console shows it).

{
  "name": "summarise",
  "base": "debian-12-20260801-72c15f01",
  "setup": [],
  "provider": "prv_8c1f04d2a7b93e650f1a2c47@latest",
  "system_prompt": "Read every file in /workspace/in. Write a one-page summary to /workspace/out/summary.md. Then call submit_result.",
  "tools": ["bash", "read", "write"],
  "custom_tools": [],
  "input_manifest": [],
  "output_manifest": [{ "path": "summary.md", "description": "The summary", "max_bytes": 100000 }],
  "environment": [],
  "size": "s",
  "timeout_minutes": 15
}

1. Publish it and run it:

jinn publish function.json
jinn run summarise --prompt "Summarise these notes for a new team member." --in ./notes --out ./result

The CLI prints what the agent does. When the run ends, ./result/summary.md holds the summary.

What to read next

• Functions: every field of a definition, bases and setup, providers, tools and secrets.
• Runs: states, failures, input and output folders, logs and limits.
• API: keys and every route. The OpenAPI document describes the same routes.
• Webhooks: get the result posted to you, and check its signature.
• CLI, SDKs and Console.
• Examples: functions for common work.

Every page is also Markdown: add .md to its address, or use the switch at the top. llms.txt lists every page for agents.