One Model Agreeing With Itself Is Not a Review

Description and use

mix2 merges two LLMs into a single team, automatically performing independent analysis and delivering a consensus answer, boosting confidence in decisions.

Problem

Developers often rely on a single LLM for code review or design decisions, risking bias and lack of independent verification.

Outcome

mix2 delivers independent, parallel LLM insights in a single terminal session, reducing manual effort and increasing confidence in decisions.

Workflow

• Install mix2 via curl
• Run mix2 in terminal
• Ask a question (e.g., code review, architecture decision)
• mix2 spawns two agents (default Claude Code + Codex)
• Each agent independently investigates the repo
• Agents compare notes and reconcile their answers
• mix2 outputs a single answer signed by the team
• If a plan is needed, mix2 generates a .mix2/<topic>-plan.md and provides the exact command to execute it
• Developer follows the plan and executes changes

Evidence

mix2 is a terminal app that turns two frontier coding agents into one small engineering team.

Source

By default, there's a maximum of two consultations per turn, enforced atomically by the Rust core.

Source