Service Architecture
Four service clusters, each with its own database (or none) and its own set of third-party integrations. This is the deployment-level view — for the planning/execution workflow model, seeAgent Model →.
Agentagent/
Hono service running the autonomous execution loop — pick up a ready task, run Claude Code, ship a PR, forward metrics.
DatabaseNone — stateless
Integrations
GitHubClaude APISlackSentry
Adminadmin/
CRUD API + UI for agents, envs, crons, tools, and tokens. Owns agent provisioning and reconcile against the cluster.
DatabasePostgres (DATABASE_URL_SHIPWRIGHT_ADMIN)
Integrations
KubernetesGoogle OAuthSentry
Task Storetask-store/
The coordination layer — owns Task, PullRequest, and TaskToken state. Deduplicates work and tracks cross-agent ownership.
DatabasePostgres (DATABASE_URL_SHIPWRIGHT_TASK_STORE)
Integrations
Sentry
Metricsmetrics/
Stateless Hono dashboard — task-store-backed JSON endpoints plus a server-rendered UI. Holds no data of its own.
DatabaseNone — stateless
Integrations
Task Store (HTTP)Sentry
Want the request-flow / workflow view instead? See Agent Model →