-
-
Notifications
You must be signed in to change notification settings - Fork 6
Quick start
- Docker with Compose v2
- Git, curl, and Node.js 22+ on the deployment host
- At least one supported provider account or API key
git clone https://github.com/thibautrey/multivibe.git
cd multivibeWarning
MultiVibe controls upstream accounts that may carry paid quotas. The shipped
Compose file uses ADMIN_TOKEN=change-me as a local placeholder, and the
inference API is unrestricted when no proxy API key exists. Replace the admin
token, configure PROXY_API_KEY (or create application keys immediately),
and keep port 1455 behind a private network or authenticated TLS proxy before
exposing it outside a trusted host.
Store both initial secrets in the ignored .env file:
ADMIN_TOKEN=<long-random-admin-secret>
PROXY_API_KEY=<long-random-proxy-secret>The existing proxy-key entry already reads .env. Replace the literal admin
entry in docker-compose.yml with required interpolation so Compose cannot
start without the secret:
- ADMIN_TOKEN=${ADMIN_TOKEN:?ADMIN_TOKEN must be set in .env}Do not commit either value. For a managed deployment, use its secret-injection
mechanism instead of .env.
./scripts/deploy.shThe deployment script records the current commit identity in .env, rebuilds
and recreates the container, then waits for /health to report the same
gitSha and buildId. Use HEALTH_URL for a remote deployment:
HEALTH_URL=https://multivibe.example/health ./scripts/deploy.shAfter startup:
- Dashboard: http://localhost:1455
- Health: http://localhost:1455/health
curl -fsS http://localhost:1455/health/health is a liveness and build-identity check only; it does not validate
storage or provider access. The authenticated /v1/models request in step 4 is
the first end-to-end smoke test.
Open Accounts in the dashboard and add at least one account. OAuth and manual-key options differ by provider; see Providers and onboarding.
export MULTIVIBE_API_KEY="replace-with-your-proxy-key"
curl -H "Authorization: Bearer $MULTIVIBE_API_KEY" \
http://localhost:1455/v1/modelsPoint an OpenAI-compatible client at http://localhost:1455/v1 and use the
same key. The requested model can be a discovered provider model or an enabled
MultiVibe alias.
# Container state and recent logs
docker compose ps
docker compose logs -f --tail=200 multivibe
# Clean stop and start
docker compose stop multivibe
docker compose start multivibe
# Update to the latest fast-forward commit and rebuild
git pull --ff-only
./scripts/deploy.shdocker compose restart only restarts the existing image; it does not rebuild
new code. To roll back from a clean deployment checkout, switch to a known-good
commit, run ./scripts/deploy.sh, and switch back to your normal branch before
the next update.