What is a squad in Tallos?
A squad is a team of AI coding agents with one leader and one or more members. The leader plans the work, hands out tasks, checks the results and reports back to you. It does not implement the work itself. Members do the work, each in its own worktree, with the agent, model and effort you choose. It is multi-agent development with a supervisor you do not have to be.
The leader runs in rounds (plan, dispatch, check) until the objective is met or the round limit is reached. It asks you questions through Tallos when it needs a decision and stops at the end with a final report. You review the changes and click Accept or Discard. See the squads feature page for the full picture.
What do I need before creating my first squad?
- Tallos with at least one connected agent. Squads can use one agent for every role; two different agents make reviews more useful. See connect your agents.
- A project added to Tallos, with a clean base branch.
- One finished solo task. If you have not run the basic loop yet, do the first-task guide first; squads build on it.
- A small, well-defined objective. A feature with a clear done state, a bug with a reproduction, or a test-coverage gap.
How do I create and run my first squad, step by step?
- 1
Open Squads
Click Squads in the sidebar. The page has three parts: New squad, Running & history and Saved squads.
- 2
Write the objective and choose the project
Under What should the squad do?, describe the result in plain words, for example “Add a dark mode toggle to the settings page”. Pick the Project. Optionally name the squad and paste a Task link from GitHub, GitLab, Linear or Jira so the leader gets the ticket's context.
- 3
Pick how the squad should work
Choose a template under How should the squad work? For a first run, Build + review is the easiest to follow: one member builds, another reviews, and they go back and forth until the reviewer approves. Templates only prefill the form; you can edit everything.
- 4
Set the leader
Pick the leader's agent, model and effort. Agent's default model uses whatever that agent is configured with. Add Leader instructions if you want to steer how it runs the squad.
- 5
Set the members
For each member, describe its job (“writes the tests”), then choose agent, model, effort and How many copies. Mixing agents works well: for example Claude Code builds and Codex reviews. Use Add member for more roles.
- 6
Check the limits and the plan
Max rounds defaults to 5 and Max members at once to 4 (up to 20 and 12). Read the plan summary under the form: it states who leads, how many members work at the same time and that everything happens in a new, separate workspace.
- 7
Start the squad
Click Start squad. Click Save as squad first if you want to reuse this team later. The run opens with the leader planning.
- 8
Watch and answer the leader
The run shows each member's status (Working, Needs you, Done, Failed, Stuck) and progress: rounds, tasks started, succeeded and failed, and an estimated cost when one is available. When The leader needs your answer appears, pick an option or write your own. Stop squad halts everyone; work already done stays in its worktrees.
- 9
Review, then accept or discard
When the status turns to Ready for your review, read the Final report and the Results by worktree. The leader may mark one result as Recommended. Click View changes to read each diff, then Accept or Discard.
- 10
Ship the accepted work
After accepting, click Open to create PR on the worktree you want to ship, review its diff once more and open the pull request, or merge it yourself.
Run your first squad on a real ticket. Get Tallos.
Which squad template should I start with?
Start with Build + review for most tasks, 3 attempts, pick the best when you are unsure how to solve something, and Leader divides the work when the objective splits into independent parts. Custom starts from a blank roster.
| Template | How it works | Good first use |
|---|---|---|
| Build + review | One agent builds, another reviews; they iterate until the reviewer approves | A small feature or bug fix. See build and review |
| 3 attempts, pick the best | Three agents try the same objective in separate worktrees; the leader compares and recommends one | A tricky fix with several possible approaches. See best of three |
| Leader divides the work | The leader splits the objective into parts, gives each to a worker and combines the results | A feature with separate API, UI and test parts. See ship a feature |
| Custom | You define every member and its job | Your own workflow, once you know the others |
What happens when I accept or discard a squad's result?
Accept marks the run as accepted; the changes stay in their worktrees and you open one to create a pull request or merge it. Nothing is merged automatically. Discard marks the run as discarded and keeps the worktrees, unless you tick the option to also delete the members' worktrees, which drops their uncommitted changes.
How do I save, reuse and schedule a squad?
Click Save as squad in the form to keep the team. Saved squads show the leader, members and limits, and offer Use (fills the form), Rename, Duplicate, Schedule and Delete. Deleting a saved squad keeps its past runs.
Schedule runs the squad on a recurring basis, like an automation: every hour, every day, weekdays, every week or a custom schedule, with a project and an objective for each run. Every finished run lands in Running & history with its status, rounds and member tally. A nightly maintenance run is a good fit; see scheduled maintenance.
Troubleshooting: why won't my squad start or finish?
- “One of the chosen agents is not installed or is turned off.” Install or sign in to that agent in Settings → Connections, or pick another one.
- “Describe every member's job.” Each member needs a job description; the leader never assigns work outside it.
- Members wait instead of starting. The leader hit Max members at once and waits for one to finish. Raise the limit (up to 12) if your machine and plan can take it.
- The run stopped before finishing. The round limit was reached. The leader reports what it has; raise Max rounds (up to 20) or narrow the objective.
- The squad could not finish. Open the leader terminal to see what happened. Work already done stays in its worktrees.
Frequently asked questions
Can the leader and members use different AI agents?
Yes. Every role has its own agent, model and effort, so Claude Code can lead while Codex and Gemini CLI work as members.
How many agents can a squad run at the same time?
By default 4 members at once, and you can raise it to 12. The round limit defaults to 5 and goes up to 20.
Does a squad merge code automatically?
No. The squad stops at a final review, and even after you click Accept you open a worktree to create the pull request or merge it yourself.
How does the leader ask me questions?
The leader calls tallos squad ask from its terminal, and the question appears in the run panel. You pick an option or write your own answer.
Can I link a squad to a GitHub issue or Jira ticket?
Yes. Paste a GitHub, GitLab, Linear or Jira link in Task link and the leader receives the ticket as its source task.
Can I run a squad on a schedule?
Yes. Save the squad, then use Schedule to run it every hour, every day, on weekdays, every week or on a custom schedule.
What does the estimated cost in a squad run mean?
When shown, it is an estimate computed from the run's recorded usage, not a bill. Your actual spend is billed by each agent's vendor under your own subscription or API key.
