Workflow splits tasks across multiple LLMs using tmux and CLAUDE.md
This workflow describes how to hand off specific tasks from Claude Code to other, cheaper or more specialized LLMs like Codex or GLM. It runs multiple AI model sessions side by side inside tmux, a terminal tool that lets you manage several sessions at once.
A script called tmux-dispatch sends a given task over to another session running a different model. Once that task is done, a called handback returns the result to the original session.
The whole setup is configured through a CLAUDE.md file plus shell aliases and scripts, which lets someone switch dynamically between different or models depending on the situation. The goal is to pick the right model for the job when hardware is limited or cost matters.
Key points
- Runs multiple LLM sessions in parallel inside tmux
- tmux-dispatch script hands off specific tasks to a different model's session
- A handback skill returns completed results to the original session
- Configured via CLAUDE.md and shell aliases, allowing flexible model switching
- Useful advanced-user pattern when cost or hardware is a constraint