How to run multiple Claude Code agents in parallel

Can Claude Code run more than one agent at a time?

Yes. Each claude process is an independent session with its own context, so nothing stops you from opening several. The limits are practical, not technical: every session draws from the same subscription or API budget, and two sessions editing the same files in the same folder will overwrite each other.

Inside a single session, Claude Code can also delegate to subagents — helpers that read files in their own context window and return a summary. That's useful for research, but it's still one conversation you're steering. Running agents in parallel usually means several top-level sessions, each owning a task.

Every approach below solves some mix of three problems:

• Terminals: where each session lives and how you watch it.
• Isolation: keeping one agent from clobbering another agent's files.
• Coordination: letting agents hand work to each other instead of you copying and pasting between windows.

What is the simplest way: several terminal tabs or tmux?

Open a tab per task, cd into the project and run claude in each. With tmux you get the same thing in one window, split into panes, and the sessions survive if you close the terminal app.

tmux new-session -d -s agents -n api "claude"
tmux new-window -t agents -n ui "claude"
tmux new-window -t agents -n tests "claude"
tmux attach -t agents

This costs nothing and works today. The catch: all three sessions share one working directory. It holds up when the tasks touch different parts of the codebase (one writes tests, one edits docs, one reads logs). It breaks when two agents edit the same file, or when one runs a formatter across the repo while another is mid-change. You are also the only messenger: if the API agent changes a contract, you tell the UI agent by hand.

How do git worktrees keep parallel agents from colliding?

A git worktree is a second (third, fourth…) checkout of the same repository, in its own folder, on its own branch, sharing one .git history. Point each agent at a different worktree and their edits can only meet when you merge.

git worktree add ../app-auth -b feature/auth
git worktree add ../app-billing -b feature/billing

cd ../app-auth && claude
# in another terminal
cd ../app-billing && claude

Claude Code can do this step for you: claude --worktree feature-auth starts a session in a fresh worktree, and running it with another name in a second terminal gives you an isolated parallel session (per the official Claude Code docs, checked October 2026). The repository needs at least one commit.

• Good: real isolation, plain git, reviewable branches, easy to throw away.
• Cost: each worktree needs its own npm install, .env and dev-server port. Merging is still your job, and conflicts show up at merge time instead of edit time.
• Not solved: the agents still don't talk to each other.

Which tools manage parallel Claude Code agents for you?

Several tools wrap terminals and worktrees into one workflow. As of October 2026, from their official sites and repositories:

Be honest about the fit. If your whole job is "spin up N isolated branches, review diffs, merge", Conductor and Claude Squad are built around exactly that loop and do it with less setup. BRIX is built around a different question: what happens when the agents need to work together, and when the product they're building also has to be sold. See BRIX vs Conductor and BRIX vs Claude Squad for the detailed comparisons.

How does BRIX run Claude Code agents in parallel?

BRIX is a desktop app where every agent is a real terminal running the CLI you already have, on the subscription you already pay for. It doesn't charge for the model. What it adds is the layer between the terminals:

• Grid or canvas. In the grid, panes rearrange with every new terminal. On the canvas you place agents freely and pin notes, task lists and images next to them.
• Edges are permissions. Draw an edge between two terminals and they can read each other's output and send each other prompts with the medmaestri CLI. No edge, no access.
• A lead agent. Tell one Claude Code to recruit workers and it spawns them as new terminals already linked to it, with a briefing, then follows up on delivery.
• Squads. Ready-made teams with roles and connections already drawn — feature (architect, backend, frontend, QA), cacabug (repro, investigator, fixer), blindagem (security audit, reviewer, cleanup) — plus your own.
• Scene view. Each agent appears as a pixel-art character whose state comes from the terminal's real activity, so you can see who is waiting for you without opening every pane.

Isolation is still git's job. BRIX runs each agent in whatever folder you give it, so the pattern is: create worktrees, then start agents in them. claude --worktree works inside a BRIX terminal too, since it's a real terminal.

How do I set up a lead agent with parallel workers in BRIX?

1. 1

   Install BRIX and open your project

   Download BRIX for macOS (Apple Silicon) or Windows. The first 7 days of STARTER are free, no card. Open a terminal in your repository.

2. 2

   Create one worktree per task

   In the terminal: git worktree add ~/code/app-api -b feature/api and git worktree add ~/code/app-ui -b feature/ui. Skip this if the tasks never touch the same files.

3. 3

   Start the lead agent

   Run claude in the main checkout. This session plans, splits the work and reviews — it is the one you talk to.

4. 4

   Let the lead recruit workers

   Ask it to spawn workers, or run it yourself: medmaestri spawn --name "API" --cli claude --cwd ~/code/app-api --prompt "Implement the /invoices endpoint". Each worker is born as a terminal linked to the lead. A lead can recruit up to five workers, one level deep.

5. 5

   Coordinate through the edges

   The lead uses medmaestri ask "API" "..." to send a prompt and wait for the answer, medmaestri send to fire and forget, and medmaestri check "API" --clean to read a worker's screen. Workers can message whoever they are linked to.

6. 6

   Review and merge

   Each worktree is a normal branch: review the diff, run the tests and merge. Then git worktree remove ~/code/app-api and close the worker.

Prefer a whole team in one command? medmaestri squad feature creates the four-agent feature squad, already briefed and connected. medmaestri squad list shows the others.

What happens after the code ships?

This is where BRIX stops being a terminal manager. The same app has studios for copy, static ads, carousels and video (BRIX PRO), and every studio reads one local product base in Markdown — product.md, audience, offer, brand. The agents that built the feature can write the launch copy and the ad for it, with the context of your product instead of a blank chat. Everything stays as files on your disk. More in From code to sales and the plans.

Frequently asked questions

Keep reading

• How to orchestrate a team of Claude Code agentsOrchestrate Claude Code and Codex agents as a team: a lead agent, canvas edges as permissions, ready-made squads, a task list and a CLI to spawn, ask and check.
• How to use Claude Code and Codex togetherHow to use Claude Code and Codex together in one workspace: one writes, the other reviews, and they message each other between terminals with medmaestri.
• BRIX vs Conductor: which one should run your Claude Code agents?BRIX vs Conductor, checked October 2026: worktrees, platforms, pricing and agents. Where Conductor wins, where BRIX wins, and how to pick the right one.