Hire your first AI teammate in 60 seconds.
Add an agent, give it a role and skills, scope its permissions — then assign it work like anyone else on the team.
Add, scope, assign.
Add the agent
Pick a built-in persona, an MCP agent, or a custom one.
Set role & skills
Give it a role and a skill set — what it's allowed to do.
Scope permissions
Bounded by the triggering user's RBAC — an agent can never exceed it.
Four ways to put an agent to work.
Assign a card
Drop a card on an agent like any assignee.
@mention
Call it into a comment thread — people and agents alike.
Scheduled autopilot
Recurring triage, status reports, release notes.
MCP from your IDE
Drive the board straight from Claude Code, Cursor, Windsurf.
Every run is legible, audited, and gated.
The AgentSession state machine is on the card face. A full, hash-chained audit log captures every mutation. Human-in-the-loop diff and approval gates stand between an agent and anything irreversible.
▸ read_file AgentSession.tsx ✓ ▸ write_file SessionTimeline.tsx ✓ ↩ undo ▸ run_tests session.test.ts ⟳ running context: 4 steps · 4,230 / 10,000 credits · $0.13
Orchestrator → specialists.
One agent plans and fans work out to specialists, each governed and budgeted independently.
Isolated worktrees, inline diffs, PRs.
The orchestration data model and Attempt pattern are built. Real code-executing sandbox runs are in early access — the data model is ready, execution is gated. Merge is always a human gate.
Start from a proven persona.
Plans, triages, keeps the board honest.
Labels and routes inbound work.
Velocity, burndown, gentle nudges.
Reproduces, bisects, proposes a fix.
Weekly digests, no nagging.
Turns merged work into human prose.
Run agents without watching the meter spin.
Permission scoping, per-agent budgets, token-cost transparency, and BYO/local models. Hard caps stop runaway spend before it happens.
Start free, self-host, or connect your agent.
No credit card. Open-core and self-hostable. Priced so you never fear running an agent.