rota

Orchestrate multiple
coding harnesses in parallel

Use your agents and accounts on the same project.
rota coordinates the work, with reviews and tests before each merge to main.

An example roundOne project · separate sessions and accounts
YOUDescribe the work
BacklogIssues #12 · #13 · #14
OrchestratorYour Claude Code session/rota-orchestrateSpawns workers · runs merge gates
One issue per worker · separate worktrees
Issue #12Account Aworktree /issue-12
Queued
PRWaiting
ReviewsWaiting
Merge gateWaiting
Issue #13Account Bworktree /issue-13
Queued
PRWaiting
ReviewsWaiting
Merge gateWaiting
Issue #14Account Cworktree /issue-14
Queued
PRWaiting
ReviewsWaiting
Merge gateWaiting
Orchestrator runs the gate: reviews + tests on current main
main · 0/3 mergedRelease On your approval
YOU → Backlog

You describe the work. It becomes three issues in the project backlog.

You set the scope, answer product questions and approve releases in your agent session. The orchestrator here is your Claude Code session. It starts workers, relays decisions and runs the merge gate.

Work across tasks and accounts

You add issues to the backlog and start a round in your coding agent. That session is the orchestrator: it starts workers, brings questions back to you and coordinates reviews and merges.

Each worker takes one issue in its own worktree. Mix Claude Code, Codex and OpenCode across your accounts. Account and worker setup

Install in your project

Paste into your agent, in your project
Install rota and set it up in this project. Follow https://raw.githubusercontent.com/l4ci/rota/main/install.md

Already installed? Start your first round.

Check changes before merging

The merge gate checks required reviews and runs your configured tests on the PR combined with the current base. Failed checks send the work back. Repeated failures can pause an issue for your help.

What the gate catches depends on your tests and configuration. Read about the safeguards and their limits.

Works with

Agent harnesses
  • Claude Code
  • Codex
  • OpenCode

Mixed in one round

Forges
  • GitHub
  • GitLab
  • Local

Self-hosted GitLab too; local projects merge directly

Issue sources
  • BACKLOG.md
  • GitHub Issues
  • GitLab Issues
  • Jira
  • Linear
  • It's a Plan
Platforms
  • Linux
  • macOS
  • Windows (WSL2)

Frequently asked questions

All questions
Why use rota if my agent already runs tasks?

rota coordinates several agents across issues and accounts. It assigns isolated work, routes reviews and checks the combined code before merging. You stay in your agent session and make the product decisions. How it works

Which coding agents can I use?

Workers can run Claude Code, Codex or OpenCode. The orchestrator is also an agent session: Claude Code, Codex, OpenCode or Hermes. Harness support

What does rota cost?

rota is free and MIT licensed. You use your own agent plans and accounts. More workers use more quota; a round summary shows the cost per issue, worker and account. Costs and quota

Can a worker merge broken code?

Workers open PRs; the orchestrator merges through the gate. It checks the required reviews and runs your configured tests on the PR combined with current main. Passing tests only cover what those tests check. You can also require approval before each merge. Safeguards and limits

What happens when I hit a usage limit?

With multiple accounts configured, rota can move an issue to an available worker on another account. If none has capacity, it waits for the limit to reset. Usage limits and unattended rounds

Do I need GitHub?

No. Use a local backlog or issues from GitHub, GitLab, Jira, Linear or itsaplan. Remote PRs and merges need a configured forge; projects without a remote can merge locally. Backlog options

Do I need herdr or tmux?

herdr and tmux host the workers' terminal sessions. rota uses them to start harnesses, send instructions and read their output. Without either, Claude subagents can share the orchestrator's account while working in separate worktrees. Behind the scenes