Skip to content

Operator guide

Documentation for teams deploying and operating the KiCI orchestrator and agent on their own infrastructure. These are the customer-deployed tiers of the three-tier architecture — the orchestrator (Tier 2) handles trigger matching and job dispatch, while agents (Tier 3) clone repos and execute workflow steps.

  • KiCI environment variable reference — auto-generated catalog of the env vars shared across the orchestrator, agent, and shared logger; per-service variables are documented in each service’s configuration reference. Regenerated from each service’s Zod schema by pnpm docs:env.

The customer-deployable orchestrator is the execution brain. It connects to the KiCI Platform relay via WebSocket, receives forwarded webhooks, fetches lock files, matches triggers, and dispatches jobs to agents. Ships as a Docker image with three operating modes: platform, hybrid, and independent.

The customer-deployable agent is the execution tier. It connects to the orchestrator via WebSocket, receives job dispatches, clones repositories, and runs workflow steps. Ships as a Docker image with label-based job routing.

How KiCI packages are distributed and deployed. Covers all three distribution channels (npm packages, OCI container images, Firecracker rootfs), orchestrator deployment modes (container, systemd, launchd, Windows service), agent deployment formats, and agent runtime dependencies.

Operator diagnostics for runtime failures that aren’t covered elsewhere. Currently documents the SDK bundle drift diagnostic — a 3-way hash compare (agent / orchestrator / host-published SDK) that collapses the Lock file is out of date investigation from hours to a single log-grep.