Fleet mode runs multiple agents in parallel on the same feature spec. Each agent works in an isolated Git worktree. Aigon’s CLI sets up and launches the implementer sessions for you, then you compare the results and merge the winner.

When to use

• You want multiple perspectives on a non-trivial problem
• The feature is important enough to warrant comparison
• You want to discover the best approach through competition

How it works

Full workflow

All commands below work as both CLI (aigon feature-start ...) and agent commands (/aigon:feature-start ..., $aigon-feature-start ..., etc.). We show CLI form here but this workflow can be driven from the dashboard and all agents supported as well.

1. Create and prioritise

aigon feature-create dark-mode
# → Describe the feature: dark/light mode toggle, CSS custom properties,
#   system preference detection, localStorage persistence...
 
aigon feature-prioritise dark-mode       # Assigns ID, e.g. #108

2. Start Fleet

aigon feature-start 108 cc gg cx

This does everything in one step:

• Creates three isolated worktrees (../<repo>-worktrees/feature-108-cc-dark-mode, etc.)
• Creates a tmux session for each agent (aigon-f108-cc, aigon-f108-gg, aigon-f108-cx)
• Opens terminal windows and launches each implementer with feature-do via the CLI
• Each agent reads the same spec but builds independently in their own worktree

The agents start implementing immediately. You can monitor progress in the dashboard or attach to any session:

tmux attach -t aigon-f108-cc     # Attach to Claude's session

If a session ends or crashes, re-open it:

aigon feature-open 108 cc        # Restart a specific agent

3. Cross-agent review (optional)

In each worktree, open a session with a different agent to review the code:

aigon feature-code-review 108

4. Evaluate

Back in the main repo, choose which agent should perform the evaluation. This is an implementation of “LLM as judge”. Use --agent=<id> when you want the CLI to launch a specific evaluator agent (the default is cc):

aigon feature-eval 108
aigon feature-eval 108 --agent=gg

Examples:

/aigon:feature-eval 108
$aigon-feature-eval 108

This moves the feature to 04-in-evaluation/ and creates a comparison template. Any agent can act as the evaluator; choose the one you want judging the implementations. aigon feature-eval 108 auto-selects cc as the evaluator, while aigon feature-eval 108 --agent=gg forces a specific one. Aigon warns if the evaluator shares a provider family with an implementer (use --allow-same-model-judge to suppress).

5. Merge the winner

aigon feature-close 108 cx              # Merge Codex's implementation
aigon feature-close 108 cx --adopt all  # Merge + review diffs from losers

The --adopt flag prints diffs from losing agents after merging, so you can cherry-pick valuable improvements (extra tests, error handling, edge cases).

6. Clean up

aigon feature-cleanup 108               # Remove losing worktrees and branches
aigon feature-cleanup 108 --push        # Push branches to origin first

Mismatch protection

If --agent=gg is specified inside a feature-108-cc-* worktree, Aigon exits with an error instead of launching the wrong agent.