Configuration basics

config.json controls everything: which reviewers run, how they’re invoked, where to put the output. It lives at the repository root (or wherever you point --config at).

Minimum viable config

{
  "reviewers": [
    {
      "id": "claude",
      "enabled": true,
      "command": "claude",
      "args": ["-p", "{prompt}", "--output-format", "json", "--allowedTools", "Read"],
      "prompt_via": "arg",
      "model": null,
      "result_path": "result",
      "timeout": 300,
      "paid": false
    }
  ],
  "synthesizer": {
    "reviewer_id": "claude",
    "prompt_file": "config/synthesis.md"
  },
  "rubric_file": "config/rubric.md",
  "output_dir": ".aireview/runs",
  "max_parallel": 3
}

What the fields mean

• reviewers[] — each reviewer is a command template. {prompt} and {model} are substituted at run time.
• prompt_via — "arg" to substitute into args, "stdin" to pipe in.
• result_path — dot-path into a JSON stdout envelope. null for plain-text CLIs.
• paid — drives a warning at checkpoint 1. Set it honestly; false for subscription/local, true for per-token.
• synthesizer.reviewer_id — which reviewer synthesizes the final review. Must match an existing reviewer.
• rubric_file / synthesizer.prompt_file / output_dir — paths are resolved relative to the config file’s directory, not the current working directory.

Caution

At least one reviewer must be enabled or the loader will refuse the config. The synthesizer must reference one of the configured reviewers (it’s reused for the synthesis call).

See the full config reference for every field and every validation rule.