Deployment Planned
Planned
The deploy kit does not exist yet. These pages describe the planned setup so you can see where chamberu is going. With @chamberu/core today you run the engine in your own Node process, start runs with dispatch() and shut down with drain().
The target workflow: agents live in their own GitHub repository, CI builds a Docker image, and Flux deploys it to your Kubernetes cluster. Every agent repository is an app with one or more agents, its own store connection, its own GitHub App and its own notifiers.
agent repo --push--> CI --image--> registry --Flux image automation--> HelmRelease --> cluster
|
OpenClaw --MCP--> app (/mcp) --hooks--> OpenClaw
GitHub ---webhooks---> app (/webhooks/github)Pieces
| Piece | Planned as |
|---|---|
chamberu package | createApp({ agents, store, notifiers, ... }) boots config, store, engine, scheduler and HTTP in order and tears them down in reverse, draining on SIGTERM. A CLI with dev, build, start, run-job and migrate |
| Base image | A slim Node 24 runtime image; agent repositories use a two-stage Dockerfile |
| Helm chart | One shared OCI chart for every agent app; each repository gets a small HelmRelease |
| Template repository | A GitHub template with an example agent, Dockerfile, CI that builds and tags images for Flux, and example Flux manifests |
Configuration
Planned configuration comes from environment variables, typically from Kubernetes Secrets: the database URL, RESUME_SECRET, the HTTP port, provider API keys (direct providers through the AI SDK provider registry, no gateway by default), static API tokens for machine clients, and OpenClaw and GitHub settings. How the Secrets get into the cluster (plain Secrets, SOPS, External Secrets) is the cluster's choice.
Observability
Planned: OpenTelemetry spans for runs and steps next to the AI SDK's own telemetry, Prometheus metrics from the HTTP app, and structured JSON logs.