What is a squad in Tallos?
A squad is a team of coding agents with one leader. The leader is an agent that plans the work, hands out tasks, checks the results and reports back to you. The members are agents that do the work, each following a job you write in plain words, such as "writes the tests" or "reviews for security".
Squads sit on top of the same pieces Tallos already uses for parallel workspaces: every member works in its own git worktree, so members never edit the same files at the same time. The difference is who coordinates. With plain parallel agents, you do. With a squad, the leader does, and you step in for questions and for the final decision.
How does a squad run, step by step?
A squad runs as a loop of rounds. One round is plan, dispatch, review. Open Squads in the sidebar, then:
- 1
Describe the objective
Write the result you want in plain words and choose the project. Optionally paste a task link from GitHub, GitLab, Linear or Jira; the leader reads it as the source task.
- 2
Pick a template
Choose how the squad should work. The template only prefills the team and the leader instructions; you can edit everything before you start.
- 3
Set up the team
Pick the leader's agent, model and effort, then each member's agent, model, effort, job and "How many" copies can run.
- 4
Start the squad
Tallos shows a plain summary of the plan, then starts the leader in a new, separate workspace. Members get their own child worktrees.
- 5
Watch or walk away
The run page shows each member's status (Working, Needs you, Done, Failed, Stuck) and live counts for rounds, started, succeeded and failed members. Open any member's terminal to look closer, or stop the squad at any time.
- 6
Review and decide
When the leader reports, read the final report, open each worktree's changes, then accept or discard.
Which squad template should I pick?
Pick the template that matches the shape of the work. Tallos ships four:
| Template | How it works | Best for |
|---|---|---|
| Build + review | One agent does the work, another reviews it. They go back and forth until the reviewer approves. | Refactors, migrations, risky changes |
| 3 attempts, pick the best | Three agents try the same objective separately, each in its own worktree. The leader compares the results and recommends one, explaining why. | Performance work, hard bugs, open-ended design |
| Leader divides the work | The leader splits the objective into independent parts, hands each part to a worker, checks each result and combines everything. | Features, test sweeps, large changes |
| Custom | Start from scratch: choose each member and describe what it should do. | Your own process |
Each template has a use-case walkthrough: build and review, best of three and ship a feature end to end.
How do I set up the leader and the members?
Every seat on the team is configured the same way, so you can match each job to the agent and model you trust for it.
- Agent: any installed, enabled agent CLI. Tallos warns you before start if a chosen agent is missing or turned off.
- Model: pick one from the agent's list, type a model id, or leave it on the agent's default model.
- Effort: the reasoning effort for that model, when the agent offers it.
- Job: what this member does. The leader must respect it and never gives a member work outside its job.
- How many: how many copies of this member the leader may run in parallel, for example four test writers.
- Leader instructions (optional): how the leader should run the squad. Templates prefill them.
The agents you can pick come from agent connections, where Tallos installs each CLI and signs it in with the subscription or API key you already have.
How do limits keep a squad under control?
Two limits bound every run: how many rounds the leader may run and how many members may work at the same time.
| Limit | Default | Hard maximum |
|---|---|---|
| Max rounds (plan, dispatch, review) | 5 | 20 |
| Max members at once | 4 | 12 |
When does the squad stop to ask me something?
The squad asks you only when the leader is truly blocked. The leader calls tallos squad ask with a question and, optionally, a few choices. The run changes to Needs your answer and the question appears under Questions from the leader.
Pick one of the options or write your own answer, then send it. The leader waits for your answer and continues from there. Answered questions stay listed on the run so you can see what you decided and when.
Ready to run your first squad?
The quickest way in is the step-by-step guide: your first squad walks through a Build + review run on a real repo.
Download Tallos and build your first squad.
What happens when a squad finishes?
When the work is done, the leader writes a final report with tallos squad report: what was done, which result it recommends and where it lives, and the open risks you should check. The run moves to Ready for your review. Nothing is merged until you decide.
- Final report: the leader's summary, rendered on the run page.
- Results by worktree: one row per worktree the squad produced, with the member's task and a View changes button.
- Recommended: the result the leader picked, marked so you can review it first.
- Open to create PR: opens that worktree so you can commit, push and open a pull request.
- Accept: marks the run as accepted. You then open a worktree's changes to create a PR or merge it.
- Discard: marks the run as discarded. Worktrees are kept unless you also choose to delete the members' worktrees.
The leader is told never to merge, push or touch your base branch.
Can I save, reuse and schedule a squad?
Yes. Click Save as squad in the composer and the team, template, instructions and limits become a saved squad.
| Where | What you can do |
|---|---|
| Saved squads | Use (fills the form), Rename, Duplicate, Delete, Schedule. Deleting keeps past runs; only the saved definition goes. |
| Schedules | Every hour, every day, weekdays or every week at the time you pick. The schedule is saved as an automation and each run starts in a new workspace. |
| Running & history | Active runs first, then past runs filtered by status, each with rounds, members started, succeeded and failed, and an estimated cost when the model's price is known. |
A scheduled squad pairs well with routine jobs, like the scheduled maintenance use case: a dependency upgrade with a reviewer, waiting for you every morning.
How is a squad different from running agents in parallel?
Parallel agents
- You open each workspace and start each agent
- You split the task and write each prompt
- You check each result and send fixes back
- Best when tasks are already independent
A squad
- The leader starts members in their own worktrees
- The leader splits the objective by each member's job
- The leader reviews and loops, within your limits
- Best when the work needs planning and checking
Both live side by side in Tallos. For the manual approach and when it wins, read multi-agent development.
Frequently asked questions
Can the leader and the members use different agents and models?
Yes. The leader and every member each pick their own agent, model and effort, so one squad can mix Claude Code, Codex, Gemini CLI, OpenCode and other CLIs you have installed.
How many agents can a squad run at once?
By default a squad runs up to 4 members at the same time and up to 5 rounds. You can raise that to 12 members at once and 20 rounds.
Does a squad merge code into my main branch?
No. Every member works in its own worktree and the leader is told never to merge or push. You accept or discard at the end, then open a worktree to create a PR or merge it yourself.
What happens to the work if I stop a squad?
The leader and its members stop working. Changes already made stay in their worktrees, so you can still review them.
Can a squad start from a GitHub, Linear or Jira ticket?
Yes. Paste a GitHub, GitLab, Linear or Jira link in the task link field and the leader reads it as the source task for the objective.
Can I run the same squad every day?
Yes. Save the squad, then choose Schedule: every hour, every day, weekdays or every week at a set time. Each scheduled run starts in a new workspace.
Do squads need a new AI subscription?
No. Squads run the agent CLIs you already have, signed in with your own subscriptions or API keys. Tallos does not resell model access.
