ONCUE / GUIDE

Getting started

Start with a repository and a small task. Get familiar with the workflow before connecting live services.

Where the project stands

OnCue is being prepared as a self-hosted, open-source project. You deploy the stack in your own accounts. The source repository currently requires access while the public release is being prepared. There is no managed signup or paid plan on this website.

Explore the local development stack

For collaborators with source access, the repository uses Node.js 24 or newer and pnpm 10.24.0. Its default development command runs local service doubles, so a first look does not need live model or compute credentials.

git clone https://github.com/ry4nrodriguez/oncue.git
cd oncue
corepack pnpm install
corepack pnpm dev

Open the web address printed by the launcher. The development sign-in route is /dev/sign-in. Local fixtures are for exploring and testing the interface; their results do not represent a live agent run.

Deploy your own stack

  1. Configure your Cloudflare account for the web app and control plane: Workers, Durable Objects, D1, and R2. Deployment uses a Cloudflare API token and account ID, or a Wrangler login.
  2. Create your GitHub App and Clerk application, with callback and webhook URLs for your deployment.
  3. Deploy the sandbox controller to Modal. Connect your Modal workspace token pair and prepare the sandbox image.
  4. Supply your OpenAI or Anthropic API keys, then enable the harnesses you intend to use.
  5. Configure the application secrets, database migrations, and repository commands described in the source deployment record before running a real task.

Live model and compute usage can incur provider charges. The project has separate development modes for live services; follow the source instructions for the exact environment and credentials.

Your first useful task

Choose something you can review in a few minutes: improve an empty state, add a focused regression test, or fix a small layout issue. Include the expected behavior, the relevant files, and the checks you want run.

Then follow the session lifecycle and inspect the result. A successful agent response is a starting point for review.

Based on application source revision 7195292 · Reviewed October 2026
Read the source architecture record (repository access required)