Using the board
Manage runs and runtimes
Follow background work in Runs. Each run has a log and can be stopped; each agent can use its own coding tool and model.
Watch or recover a run
Open Runs or ask your coding agent "what's running?". The app and akb run show the same
runs.
| You want | Action |
|---|---|
| Read progress | Open the run's log; use --follow in the terminal to watch it. |
| Stop work | Stop the run; this also ends commands it started and releases the card. |
| Continue failed work | Use Resume where supported. |
| Start the same request again | Use Retry, including when an early failure cannot resume. |
For interrupted card creation, choose Resume creating or Discard on the board. Discard archives the unfinished card as Rejected.
Runs, logs and chats stay on your machine, outside the repository. Moving or renaming the project starts an empty run history.
Sub-runs
An agent can delegate part of its work with akb run start "<task>". In Runs, a sub-run
appears under its starting step with its own log, cost and model. Its changes land with the
parent's; stopping the parent stops its sub-runs.
Which runtime each agent runs on
A runtime specifies a coding tool, provider, model, reasoning effort and connection settings. Pick it on the agent's page under Configuration → Workflows or Configuration → Board.
Manage runtimes under Configuration → Runtimes. Global default supplies agents without a runtime of their own. A workflow's planner uses Planning helper's runtime until you assign one.
akb agent
akb agent runtime add "My gateway" claude-code
akb agent set --runtime my_gateway baseUrl https://my-gateway.example.com
akb agent bind builder my_gateway
Use Test or akb agent test to send a small test message. Enter API keys yourself in settings
or the CLI; do not paste them into chat. Runtimes and keys are machine settings, shared by neither
git nor a Cloud workspace.
See Multi-runtime support for supported tools, their settings and execution limits. A single chat can use a different agent or model.