Working with Multiple Coding AI Agents: Just Use tmux

Note#195

notes
Published

2026-01-10

When you want Claude Code to drive Codex, or coordinate multiple agents (via OpenCode, Amp, or pi), don’t over-engineer it with hooks or whatnot. Just use tmux.

Use tmux as a Bridge

The trick is to treat tmux as a coordination layer. Have one agent create a new session, window, or pane that runs another agent CLI, then drive it with tmux send-keys. That gives you a simple and scriptable way to coordinate multiple agents.

Example

# Start a second agent in its own tmux session
tmux new-session -d -s codex 'codex'

# Send instructions
tmux send-keys -t codex -l 'Read /tmp/instructions.md and follow it.'
tmux send-keys -t codex Enter

# Inspect output
tmux capture-pane -t codex -p -S -100

Practical Caveats

You can also target panes using <session>:<window>.<pane> (for example codex-review:1.2).

Sending Enter can be flaky if you include newlines in the instructions string. I’ve had better results when I explicitly instruct the agent to press Enter after sending the instructions, and when the actual tmux send-keys ... Enter is its own command. Consider documenting this in your AGENTS.md or an agent skill.