Mel – the agentic terminal
The AI terminal for teams.
Learn · parallel AI coding agents
How to run multiple AI coding agents in parallel
Updated
To run several AI coding agents in parallel, give each agent its own git worktree and its own terminal pane, so no two agents edit the same files. Or hand one big task to a team of agents, where a lead splits the work and workers build their parts at the same time.
Mel – the agentic terminal does both, in one window. Parallel agents are the fastest way through a backlog, and the fastest way to make a mess, if two of them write to the same folder. The rules below keep the speed and lose the mess.
Two kinds of parallel
Separate tasks, side by side. A bug fix in one agent, a docs update in another, a new endpoint in a third. Each agent works alone; you are the one who merges.
One task, split. A feature that touches the API, the tests and the docs, split across workers by a lead agent that plans the parts, checks they fit and reports back.
What breaks when agents run in parallel
- Shared files. Two agents editing one file undo each other. Fix: one worktree per agent, or a lead that gives each worker its own files.
- Shared ports and databases. Two dev servers on port 3000 collide. Fix: a port per worktree.
- Lost track. One agent waits for an approval in a hidden tab for an hour. Fix: a terminal that tells you when an agent is blocked.
- Unreviewable output. Five agents produce five big diffs. Fix: small tasks with a clear finish line, reviewed one at a time.
- Cost. Every agent's steps use model tokens. Fix: run in parallel only when the work is truly independent.
How Mel does it
- A team of agents on one task. Type
/orchestrate-agentand describe the task. A lead plans the split and workers open in their own panes, running in parallel. You see the proposed team on a confirmation card before any worker starts. - Claude Code or Codex as workers.
/orchestrate-climakes the workers the CLI agents you already use, running in Mel panes. - A pill for every worker. While a team runs, a strip shows the lead and each worker with its status, model and tokens; a click opens that worker's session.
- Separate tasks, side by side. Split a pane with ctrl+D and start an agent in each, or use + → Worktree for one checkout per agent.
- Know who needs you. Blocked agents show in the sidebar and the inbox, and the Runs screen lists every run that needs you.
- Review in one place. The code review panel shows the combined changes.
Parallel agents, compared
| Several terminal windows | Mel, side by side | Mel, team of agents | |
|---|---|---|---|
| Who splits the work | You | You | The lead agent |
| Isolation | Your own worktrees | + → Worktree | The lead gives each worker its own files |
| See which agent is blocked | Check each window | Sidebar, inbox, Runs screen | Worker pills, inbox, Runs screen |
| Who checks the combined result | You | You | The lead, then you |
| Best for | A few unrelated jobs | Several unrelated jobs | One many-step job |
FAQ
Can I run several AI coding agents on one repository? Yes, if each has its own worktree or its own files. Two uncoordinated agents in one folder will overwrite each other.
Is a team of agents faster than one agent? For many-step work with independent parts, often yes. For a one-line fix, one agent is faster and uses less.
Can I run Claude Code and Codex in parallel in Mel? Yes. Start each in its own pane, or let a Mel lead start them as workers with /orchestrate-cli.
Does running agents in parallel cost more? Yes. Every agent's steps use model tokens, so check usage after your first parallel run.
Claude Code and Codex are trademarks of their owners. Mel is not affiliated with or endorsed by Claude Code, Codex, Anthropic or OpenAI.
Mel – the agentic terminal · openmel.dev. A multi-agent, multi-model AI terminal for Mac, Windows, Linux and VS Code. Free with your own API key.