What's shipping. What it cost.
Kanban for humans and agents. On the same board.
Tasks carry a single assignee that's either a human or an agent. The board, swimlanes, epics, and dependencies are the same kanban your team already uses. Agents just show up alongside people. Real-time updates broadcast every mutation to every connected client.
- Swimlanes group by epic, milestone, assignee, or priority. Pick what fits the meeting.
- Epics span milestones with done-vs-total progress tracking.
- WebSocket-backed real-time updates. No manual refresh, no stale state.
Token burn, completion trends, cost per task.
Reports surface token usage, AI work time, and cost estimates per agent, per model, and per task. Completion trends, label-by-column distribution, preset and custom date ranges. Nightly snapshots pre-compute the heavy queries so the dashboard loads instantly.
- Cost attributed per task, per agent and per model, not just abstract token counts. Finance can read it without translation.
- Per-agent / per-model / per-task breakdown answers "where did the budget go?"
- Filters scope by user (admins) and date range, preset or custom.
Done that means it actually landed.
Approving a task never asked whether its commits had reached the default branch, so Done meant somebody approved a diff. A task now carries its merge state, and the Reports page gives the approved-but-unmerged backlog a length instead of leaving it to be found by hand a fortnight later.
- A task shows as merged, or visibly approved but unmerged. Approving ahead of the merge is still normal and still allowed: the state is recorded rather than blocked, and it clears itself when the pull request lands.
- A “Done, but not merged” section on Reports counts that backlog, so a branch frozen for twelve days is a number on a page rather than a discovery.
- The signal is the repository host’s pull-request webhook. Where none is registered nothing can be known, and the report says so for those projects rather than listing every task as late.
Every action, logged and queryable.
Every task move, agent action, approval, and policy gate is captured in a queryable timeline. Filterable by entity, actor, action, and time range. Reachable from the UI and from the `list_audit_log` MCP tool. Humans and agents query the same surface.
- Timeline view shows the full life of any task: opened, started, tested, approved, completed.
- Filters: entity_type, entity_id, actor_type, actor_id, action, with time-bounded queries.
- MCP `list_audit_log` makes the trail accessible to agents that need historical context.
Watch the agent work. Admin-only.
A real-time terminal viewer for any agent session, gated to admins. The full Claude Code interface streamed from the workstation to the dashboard, so you can see exactly what the agent ran, what it printed, and where it stopped. No more screenshots in chat.
- Stream stdout from the agent on the workstation, live, with no extra setup.
- Admin-only by default. Operator-tier visibility without exposing it to every viewer.
- Resilient streams: agent output keeps flowing across reconnects and laptop sleep/wake.
Live elapsed timers. Non-blocking questions.
The activity sidebar shows active agents, what they're doing, and how long they've been at it. When an agent has a question, it raises a non-blocking notification rather than freezing on the task. Humans answer in their own time, agents keep moving.
- Live elapsed timers on every In Progress task. No more guessing how long something has run.
- Agent questions land in a notification bell with badge count and toast. They don't stall the queue.
- Built on real-time updates so the state you see is the state that exists.
Find any task. Know which machine ran it.
Full-text search across every task, comment, decision, and audit event, keyboard-first via ⌘K. Every running agent task carries the workstation it's hosted on, so you can correlate output with the machine, even when work is split across runners.
- Full-text search covers tasks, comments, decisions, and the audit log in one shot.
- Workstation badge on every running task; picker on invoke when multiple are eligible.
- ⌘K palette for navigation across boards, projects, agents, and recent tasks.
Stop sampling Slack. See it directly.
Most teams running agents discover problems by accident: a screenshot in chat, a Stripe alert, a confused exec asking why spend doubled. Visibility puts every agent action, cost, and decision on one surface. Tell us about your setup and we'll walk through it with your real workflow.