Skip to main content

Model Peer

Cross-model peer review for coding agents.

Model Peer lets Claude Code, OpenAI Codex CLI, and Google Gemini CLI consult one another as independent, read-only engineering peers.

model-peer ask claude "What edge cases am I missing?"
model-peer ask codex "Review this authentication design"
model-peer ask gemini "Challenge this migration plan"

model-peer review

There is no chat between models​

This is the part that surprises people. Model Peer is not a conversation between models — your agent is the hub. Each consultation spawns another vendor's CLI, read-only, gets one answer, and exits. Nothing persists.

Primary agent
    |
    +--> independent peer model
    |        |
    |        +--> advisory response
    |
    +--> primary agent evaluates the advice

The peer supplies evidence, not authority. Project rules and invariants still win.

You do not paste code into the question​

The peer starts in your current working directory with read tools enabled. Name files and symbols and let it look for itself:

$ model-peer ask codex "What does stack_contains in bin/model-peer guard against?"
# codex reads bin/model-peer itself, then answers

Why​

Coding agents can review their own work, but self-review is still self-review. Model Peer makes it trivial to ask a different model:

What am I missing?

model-peer review goes one step further. Every available reviewer receives the same Git status and patch independently, without seeing the other models' conclusions. Only after all reviews finish does a synthesizer reconcile the findings.

              +--> Claude --+
              |             |
git changes --+--> Codex ---+--> synthesis
              |             |
              +--> Gemini --+

Because reviewers cannot anchor on each other, agreement between them is real signal. The synthesizer is explicitly instructed not to accept a claim merely because several models repeated it.

Supported CLIs​

ProviderCommandConsultation safety
Anthropic Claude CodeclaudePlan mode, read-only inspection tools, stdin closed
OpenAI Codex CLIcodex--sandbox read-only, ephemeral session, stdin closed
Google Gemini CLIgeminiPlan mode, explicit deny policy, extensions disabled, stdin closed

Model Peer does not collect or store credentials. Authentication stays with each official CLI.

Design principles​

  1. Independent review — reviewers start from the same evidence, not each other's opinions.
  2. Read-only consultation — asking for advice should not hand over write authority.
  3. Primary-agent ownership — peer output is evidence, not a command.
  4. Project rules win — local invariants outrank generic model advice.
  5. No credential handling — authentication stays with official vendor CLIs.
  6. Bounded chains — depth is capped and opt-in, and no model ever consults itself.
  7. Review before autonomy — v0.4.0 analyzes; it does not automatically apply fixes.

Next​