Everything between
"run the agent" and "merge"
The orchestration, review, and observability layer for agent-written code — without building bespoke plumbing.
The full transcript, streaming
Reasoning, tool calls, file edits, inline diffs, check output, elapsed time, tokens, and cost — streamed into your editor's sidebar and this dashboard while the run is still going.
- Follow along or jump in when something looks off.
- Reconnects resume exactly where the stream left off.
- Per-phase model, effort, and persona chips.
Nothing advances unreviewed
Each gated phase parks the run for a human decision. Approve it forward, or send it back — your note becomes the agent's rework instruction on the same branch.
- Approve from the editor or the web dashboard.
approvers:policies name who may approve, by their GitHub or Linear identity.- Native code review in the editor submits comments as one "request changes".
Built for teams, not demos
Declarative workflows
Phases and typed steps — prompt, check,
command, git, notify — in YAML
that lives in your repo. Change the pipeline in a pull request.
Task board & statuses
User-defined columns drive execution: a phase starts when a task
enters its onStatus column and moves it on with
setStatus when it lands.
Trackers
GitHub Issues & Projects, Linear, Jira, GitLab, and Notion sync into one backlog — native statuses, both directions, with routing rules per label or assignee.
Notifiers
Slack, Discord, Microsoft Teams, Telegram, Google Chat, Mattermost, or any webhook — gate arrivals, results, and published artifacts land where your team already looks.
Runners & isolation
Local or remote runners, selected by labels. Every run executes in a fresh Docker or gVisor sandbox with repo-scoped, short-lived credentials.
Personas
Reusable agent identities — name, icon, color, system prompt — assignable per step. Your architect and your implementer stop sounding like the same agent.