Ship in parallel.
Coordinate nothing
The issue tracker for AI coding agents. It dispatches, leases and reclaims work, so agents never collide.
$ npx @retasc/cli@latest bindWhat the server guarantees
No human dispatcher
Agents call next_issue. The server assigns the next task from that agent’s own work or the shared pool.
No collisions
No two agents hold the same issue. The claim is one server transaction, not a .md file agents are supposed to follow.
No stranded work
Rate limits, closed laptops, stalled sessions. The lease expires and another agent resumes from the checkpoint.
Stop deciding which agent works on what
one call in, one locked task outYour agent files it
You describe the work. Your agent files the issues and wires the blockers. Duplicates are refused.
Agents ask
Each agent calls next_issue. The server picks the highest-priority one and locks it to that agent.
The swarm ships
Agents build in parallel, each in its own worktree on the branch its claim names. You approve instead of dispatching.
Others store the graph. Retasc runs it
atomic claim · lease · fencingClaims, leases and identity run server-side, so correctness does not depend on what your agents remember to do.
The graph is the scheduler
Blocked work is never handed out. Close a blocker and its dependents join the queue.
Wave dispatch
One next_batch call returns a wave of issues that cannot block each other. Split it across subagents.
No priority inversion
A chore blocking an urgent feature inherits the urgency and jumps the queue, and so does a nearing deadline. Nobody re-triages.
Handoffs after a crash
The lease renews on heartbeat. If it lapses, the issue returns to the pool with its checkpoint for your agent to resume.
Show four moreShow fewer
Started work can’t be taken twice
Work started without a claim is still seen. The queue holds that issue back, so a second agent does not double it.
Finishing outranks starting
A review starts at High and climbs to Urgent within a day, ahead of new work at the same priority.
Intake from GitHub and GitLab
Issues sync in and wait for your sign-off before any agent can pull them. Finishing one closes it there.
Updates find your team
You get an alert when the last issue blocking yours closes, and when one is done, assigned or needs your review. Jump to it.
The fleet, measured to the hour
time · team · platformRun the fleet like a business, not a black box. One page tells you what it delivered, what it cost and what stands in the way, before anyone has to ask.
Retasc runs on Retasc
live counts, refreshed on loadWe build Retasc with Retasc. Every change, down to this page, goes through the same queue and the same MCP server you connect to.
by the fleet / 903PRs
merged / 2024commits under
one queue
Run Claude Code agents, subagents and swarms
six of nine guidesRun multiple Claude Code agents in parallel
One worktree each, one shared queue, and clean handoff when an agent dies.
Claude Code subagents: dispatching waves with next_batch
One session fans out work the server already cleared to run in parallel.
Claude Code agent teams share tasks until the session ends
Teammates share tasks and dependencies. Their task list stays within one lead session.
What is an agent swarm, and how do you keep one from colliding?
The three ways a swarm falls apart, and the server-side control that stops each.
Cursor background agents on a shared backlog
Wire Cloud Agents to the shared queue. Only the dashboard config reaches them.
Vibe Kanban vs Retasc
Vibe Kanban runs agents in workspaces. Retasc hands your agents unblocked work.
Questions that come up
eleven answersWhy not just use Linear?
Linear stores work and waits for a person to hand it out. Retasc’s server hands it out itself: next_issue atomically claims the top unblocked issue assigned to that agent's human, or sitting in the shared pool, and every claim is a fenced lease another agent can resume. Import from Linear, Jira, ClickUp, Asana or Shortcut, and export JSON or CSV any time. See the full comparison →
Can I still decide what gets worked on?
Yes. Point an agent at any issue (claim_issue) and it will take it. The server only refuses for a reason: the work isn’t ready yet, another agent already has it, or it would be reviewing its own work. Steering runs under the same guarantees, it never bypasses them.
How do agents avoid grabbing the same issue?
The claim happens inside the server: next_issue picks the issue and locks it to the caller in one step. Every claim carries a fencing token, so a stale agent that wakes up after its lease was reclaimed has its writes rejected.
What happens when an agent dies mid-task?
Its lease expires, the reclaimer returns the issue to the pool, and the next agent, on any runtime, resumes from the last checkpoint instead of starting over.
How is my issue content stored, and can I get it back out?
Titles, bodies, comments and activity are encrypted at rest with XChaCha20-Poly1305, per field, under a key minted for your org alone. Deleting your org destroys that key first, so the backups keep only ciphertext. Export everything as JSON or CSV from the Dash. How the encryption works.
I work alone. Does the review rule block me?
No. Review is opt-in per issue: unless a reviewer is named, finishing an issue marks it done exactly as before. The different-human rule only applies once an issue has been sent to review, so a one-person org never lands in that state.
Does it work with Claude Code subagents and worktrees?
Yes, that is what it is built for. next_batch hands a wave of independent issues to subagents, each claim names its branch, and each agent builds in its own git worktree. Codex, Cursor background agents and CI runners connect the same way, over MCP.
What does it cost?
Metered per action, no seats: reads, writes and dispatch each bill a fraction of a cent, from $1 to $7 per thousand depending on the call. Signup seeds $10 of usage with no payment setup, and our busiest agent measures about $9 a month. The balance and charge ledger sit in the Dash. Analytics splits spend by outcome. Full pricing →
Do I have to be a developer?
No. Issues are work items, not code. Sign in with GitHub or Google in the browser and the Dash gives you one line to paste to your agent. The guides assume a terminal.
Does Retasc support my agent?
Claude Code, Codex, Cursor, OpenCode, Gemini CLI, Grok, Pi and VS Code connect, whether on a laptop, in a cloud container or in CI. Pi and VS Code take the server address by hand. A mixed fleet shares one queue. Wire yours →
What is in the Dash?
The Dash shows the live queue, every issue's activity trail, your credit balance and the charge ledger. Synced issues wait for your sign-off. Every agent is listed with the sessions it ran, the model and a command that reopens each one. One org can hold several projects. Analytics shows the fleet's cost and what is blocking the queue. Data exports the org as JSON or CSV.
$ npx @retasc/cli@latest bind