Concepts

Keelson is the local harness around coding agents. It does not replace the agent; it gives the agent a governed place to run, remember, repeat work, call tools, and surface results. The concept pages build that mental model: what keelson is, why it is shaped this way, and the vocabulary you will meet everywhere else.

Each page carries one organizing idea, and they read best in order: the positioning first, then the map, the live chat loop, the extension model, the two engines, where their work is scoped, the state behind them, and the provider underneath every turn.

What Keelson is, and is notThe positioning: a local, single-user harness, the non-goals it deliberately leaves uncovered, and the honest status of the beta.

ArchitectureThe map: what the harness owns, where ribs attach, and the one rule that keeps them apart.

ChatThe live loop: one agent turn, the tools ribs give it, the provider you choose, and the governance that gates every call.

The rib modelWhy capabilities are packages, what a rib can contribute, and the boundary the harness enforces rather than requests.

WorkflowsDeterminism next to an agent: the DAG execution model, the two data channels, and where the agency is bounded.

Projects and worktreesScoping work to a repository: how chat binds in place, how a run isolates into a git worktree, and the lifecycle of both.

Snapshots and surfacesRibs publish data, never UI: frames, the triple fail-closed pipeline, and the closed canvas vocabulary.

UsageThe read-only token ledger: per-turn usage across chat, workflows, and ribs, in six views from live pulse to raw ledger.

Memory and stateOne home, one database, one keychain, and how a memory earns the right to instruct the agent.

ProvidersThe swappable agent: one interface, several coding-agent SDKs behind it, chosen per turn and per node.

GlossaryThe vocabulary in one place, with the two overloaded terms disambiguated.