Leyline

First run ​

Trace capture starts on for new runs. Before submitting private data, check Capture new runs in the trace footer.

Start Leyline ​

If you followed Linux installation, open Leyline from your application menu and continue at step 3. The packaged app does not need a development server.

  1. Run the development server:

    bash
    npm run dev

  2. Open the Vite URL, usually http://localhost:5173/.

  3. Select a recent project, or select Add new project.

  4. Select a model and thinking level if you do not want the pi defaults.

  5. Enter the first prompt.

  6. Press Enter or select the send button.

Leyline creates the session, applies the staged runtime choices, and sends the prompt. You do not have to create an empty session first.

The optional ultrafast chip is off by default. On Home, clicks stage the choice without a confirmation dialog. It uses premium pricing for supported OpenAI Responses models. See Runtime controls for model and authentication limits.

If no usable model is available, open Settings → Models & providers. Sign in when the provider requires authentication, or add a custom endpoint and model without a key. Test connection can check saved settings or unsaved editor values before the first prompt. See Provider settings.

Check trace capture ​

Capture starts on for new agent runs. Traces can store raw prompts and tool results on the selected backend.

Before the first prompt, select the Trace explorer icon button at the bottom of the sidebar. Expand Capture new runs in the footer. To prevent capture of new runs, clear the Capture new runs checkbox. Turning capture off does not delete existing traces or interrupt capture for an active run. See Tracing.

Use the workbench ​

The workbench shows the project and session breadcrumb, transcript, live output, and composer. Thought, skill, and tool rows can expand. Select a System row to inspect recorded prompt and tool changes in the side pane.

The composer stays available during an active run:

  • Press Enter to add an editable task to Up next.
  • Press Option+Enter to steer the current run.
  • Press Shift+Enter to add a line break.
  • Select the stop button to interrupt the run and hold remaining unsent tasks.

Select the attached Up next tab to manage pending tasks. Use Send options beside the send arrow to queue or steer without keyboard shortcuts. If the queue is held, select Resume to continue queued work. See Composer for editing and queue controls.

Select Session traces to inspect a captured run. Events shows a temporary event list for the page connection, separate from durable traces.

Use Session details, Memory, Events, and Export transcript in the workbench header. Use Reload runtime at the bottom of the sidebar when pi resources or saved provider/MCP configuration must reload.