Skip to content
The Test Cabinet The Test Cabinet

Run the Local Service Stack

Overview

Section titled “Overview”

Bring the service-driven flow up on a local k3d cluster: backend, dispatcher, artifacts, repos, publisher and arena services. Runs are enqueued at the backend, and an in-cluster dispatcher schedules each one as a per-run driver Job. The console runs from source against the forwarded services.

The full walkthrough is Running the Local Service Stack; every service variable is in Running.

Prerequisites

Section titled “Prerequisites”
  • Docker (or a Docker-compatible runtime), k3d, kubectl, and make on PATH. k3d’s first-class runtime is Docker; Podman must be rootful and expose the Docker socket.
  • The microservice-*, contracts, engines, tool-* and test-suites repositories checked out under the superrepo (Cloning).
  • A harness API key exported, or set in the superrepo’s .env: ANTHROPIC_API_KEY, OPENAI_API_KEY, or OPENROUTER_API_KEY. The Makefile reads it into the cluster Secret. See Set Up Authentication.

Steps

Section titled “Steps”
Terminal window
export OPENROUTER_API_KEY=…                   # the harness you'll run


make -C deployments/local local-up            # cluster + images + secrets + overlay + ingest
make -C deployments/local local-forward       # hold the forwards open (its own terminal)
npm run dev                                   # in console.testcabinet.ai/ (its own terminal)

local-forward exposes the backend on :8787, artifacts on :8790, arena on :8791, repos on :8792, and Grafana on :3000. The console’s dev server proxies /api, /artifacts and /arena to them, so open http://127.0.0.1:1430 and the catalog loads. The local overlay verifies no Access assertion, so there is nothing to sign in to: every mutation is recorded under the identity the console sends in Cf-Access-Authenticated-User-Email, dev@local by default. Then, in the console:

  1. Enqueue a run: pick a case, a model, and the harness whose key you exported. The console streams its events live.
  2. Watch it schedule with kubectl -n tcab-local get jobs,pods -w. The finished run is reviewable, its build served from the artifacts service.

Manage

Section titled “Manage”
Terminal window
make -C deployments/local local-status        # pods, services, volumes
make -C deployments/local local-rebuild       # after a code change: rebuild + restart
make -C deployments/local run-images          # after changing run-container tooling
make -C deployments/local local-ingest        # after editing a case: force re-ingest
make -C deployments/local local-down          # delete the cluster and everything in it

local-rebuild rebuilds the long-lived service images from their checkouts. Changes to the tooling baked into the run-container images need run-images.

Next steps

Section titled “Next steps”