Install

Model Peer is a single Bash script with no runtime dependencies beyond the vendor CLIs you choose to use.

One-line install

curl -fsSL https://raw.githubusercontent.com/makedirectory/ModelPeer/v0.8.1/install.sh | bash

Interactive setup, which can offer to install missing CLIs:

curl -fsSL https://raw.githubusercontent.com/makedirectory/ModelPeer/v0.8.1/install.sh | bash -s -- --setup

As with any remote shell installer, inspect it before piping it into a shell.

From a cloned repository

git clone https://github.com/makedirectory/ModelPeer.git
cd ModelPeer
./install.sh

This installs:

~/.local/bin/model-peer
~/.local/bin/ask-claude
~/.local/bin/ask-codex
~/.local/bin/ask-gemini
~/.local/bin/ai-review

The ask-* and ai-review commands are compatibility shortcuts. The primary interface is model-peer.

Model Peer never asks you to paste an API key or password into its installer. Vendor login flows remain vendor-owned.

Override the install directory with MODEL_PEER_BIN_DIR or --bin-dir:

./install.sh --bin-dir /usr/local/bin

Install the vendor CLIs

At least one is required for ask; at least two for review.

Provider	Install	Documentation
Claude Code	Setup guide	CLI reference
OpenAI Codex CLI	Codex CLI	Docs
Gemini CLI	Get started	Docs

./install.sh --setup can offer to install missing CLIs, preferring official installation methods available on the machine, then leaving authentication to each vendor.

Authenticate

Claude Code

claude auth login

Codex

codex login

Gemini

Run Gemini interactively and choose the Google login flow:

gemini

Gemini also supports GEMINI_API_KEY and Vertex AI authentication. Model Peer does not read or persist those credentials.

Verify

model-peer doctor

doctor reports which CLIs are installed, whether authentication looks available, the safety defaults in force, the per-provider nested-consultation matrix, the effective peer-chain depth limit and the longest chain actually reachable, and which Model Peer skills the project you are standing in has installed.

It reads configuration only, so it costs nothing and can be wrong about credentials a CLI keeps somewhere it cannot see. To settle a question rather than infer an answer, model-peer doctor --probe runs one real consultation per CLI — see the reference.

Set up a project

Installing Model Peer gives you the command. It does not yet give the coding agent in a repository any reason to use it. Run this once per project:

cd ~/code/your-project
model-peer init all

init writes only the agent directories you name — all, or any of claude, codex, and gemini. With all, that installs a cross-model-review agent skill where Claude Code, Codex, and Gemini each look for one, plus a /peer-review slash command for Claude Code. It never touches your AGENTS.md, CLAUDE.md, or GEMINI.md. Commit the result and your teammates get it too.

→ In your workflow

Uninstall

./uninstall.sh

This removes Model Peer commands only. It does not uninstall vendor CLIs, touch their credentials or configuration, or remove skills from your projects — those are committed to your repositories.