From the repository root, run npm ci, then npm run examples. All examples
use deterministic runners, require no credentials, and make no model calls.
Replace a runner with the documented provider adapter for live use.
| Example | Run | Expected result |
|---|---|---|
| Design comparison | npm run demo |
Decision chooses comparison, benefits/risks run concurrently, answer synthesizes them; brief branch skips |
| Code review | npx tsx examples/code-review.ts |
Correctness and test reviews run independently; final review lists both findings |
| Failure handling | npx tsx examples/failure-handling.ts |
completed; optional failure stays in history and join completes with local findings and explicit remote error context |
| Loop and live update | npx tsx examples/execution-control.ts |
Two refinement rounds, a pause, and an atomic graph update/resume |
| Custom runner | npx tsx examples/custom-runner.ts |
Received direct predecessors: route |
These text-only examples use temporary non-Git directories. The code-review
example supplies a short diff in goal. In a real Git checkout, core manages
worktrees and merge agents; Pi can inspect and edit its assigned workspaces.
Fake responses illustrate control flow; they do not demonstrate model quality.
Export OPENAI_API_KEY and BRAID_MODEL; optionally export OPENAI_BASE_URL.
Then run npm run demo -- --live. This makes billable requests and needs a model
that supports the decision function schema. .env.example documents variables;
Braid does not automatically load .env files.
npm run demo:panel regenerates the panel snapshot from
the actual TUI renderer using deterministic fake model responses. It does not
require a running Pi session or model credentials. Install the extension and
run /braid in Pi to see real jobs update while they execute.
