Docs
Goobers documentation
Goobers is an open, self-hosted platform for running a workforce of AI coding agents against your own repository and backlog. A YAML workflow compiles to a deterministic state machine — the agent does the work inside it, but never decides what happens next.
Reference lives in the binary
goobers help all lists every command; goobers schema, explain, and examples project the DSL's closed schemas and
real examples straight from the build you're running, so they can never drift from your
version the way a page on this site could. goobers features and versions report what your build supports; help stages covers the
workflow-internal commands. The installer also writes a version-matched copy of the
product's own docs to ~/.local/share/goobers/<version> — that copy,
not this site, is the reference for the exact release you're running. The pinned upstream
tree this site links into is Agent-Clubhouse/Goobers@a512230 .
What Goobers does not do
There is no hosted Goobers, and none is planned — you run your own binary on your own
machine. The portal reads the journal and telemetry; it cannot create or edit a
definition. You bring your own coding-agent harness, and there are exactly two: copilot or claude-code. Agentic stages are not sandboxed by
default. instance.yaml is read once at startup — changing it needs a
restart — though a Git workflowSource reconciles continuously once the
daemon is up, a distinct thing from the startup read. The run journal is
tamper-evident, not tamper-proof. And the Kubernetes-and-Temporal tier is quarantined in
the product's own repository: a deployment goal, not something you can run today.
This site documents Agent-Clubhouse/Goobers@a512230.