Shape B — cloud-only multi-agent (default for a fresh install with CLIs authenticated)

.local-review.yml — example: mix CLI agents with a local Ollama provider

llms: # CLI agents — names are well-known (claude / codex / copilot / gemini). # Just set the per-agent knobs you care about; auto-detection handles the rest. claude: model: claude-opus-4-7

codex: enabled: false # opt-out (still runs if --only codex is passed)

gemini: model: gemini-2.0-flash # llms.gemini.force_after_sunset: true # uncomment to keep trying past 2026-06-18

# Provider agents — ANY entry under llms: with a base_url: becomes an # HTTP provider agent (Ollama / vLLM / OpenAI-compat). Name is free-form. ollama: base_url: http://localhost:11434/v1 model: qwen2.5-coder:7b

merge: preferred_llm: auto # or: claude, codex, copilot, ollama, ...

See [`examples/.local-review.yml`](examples/.local-review.yml) for full schema.

**Three common shapes:**

.local-review.yml — every field is optional

prompts: prepend: | Additional house rules: - Never approve commented-out code. - Flag any new dependency in package.json or go.mod. review: min_severity: warning # nit | info | warning | major | critical

You can ship the whole pack of overrides this way — see [Customise for your team](#customise-for-your-team) below.

> **Security: a repo's `.local-review.yml` is untrusted by default.** Because it ships inside code you might be reviewing for the first time, the execution / network / secret fields — `llms.<name>.cli_path` (runs a binary), `base_url` (sends your diff to a server), and `api_key` — are **ignored when they come from the repo config**, with a warning. House-rules fields (`prompts`, `review`, model/timeout) still apply. Put `cli_path` / `base_url` / `api_key` in your own `~/.local-review.yml` instead, or — if you genuinely trust the repo (e.g. a team-standardised LAN Ollama endpoint) — set `LOCAL_REVIEW_TRUST_REPO_CONFIG=1`.

**4. Review your current branch** vs `main`:

Findings print to your terminal. The tool exits non-zero on major / critical, so it slots straight into a pre-commit hook.

5. (Optional) Check how the LLMs score on a labelled benchmark. The repo ships with bench/RESULTS.md — a leaderboard (precision / recall / F1 / noise, plus uplift-over-raw-LLM and overhead-vs-raw-LLM tables) generated from a 10-case dataset spanning Go / TS / Python / Rust. Refreshed before each release; you don't need to run anything locally. See bench/README.md for methodology.

---

Restrict to a subset (overrides config)

local-review review --only claude,ollama

No config file needed — every authenticated CLI runs automatically.

Configure

local-review loads YAML from a cascade — built-in defaults → ~/.local-review.yml → ./.local-review.yml → CLI flags.

A minimal ~/.local-review.yml (unified shape, v0.14+):

llms:
  ollama:                              # any name you like — "ollama", "qwen", "cloud", ...
    base_url: http://localhost:11434/v1
    model: qwen2.5-coder:7b
review:
  min_severity: warning
  max_findings: 20

Any entry under llms: with a base_url: becomes a provider agent — an OpenAI-compatible HTTP endpoint (Ollama, vLLM, OpenAI, Anthropic, Mistral, DeepSeek, Kimi, Qwen, Together, Groq, OpenRouter). Provider agents run side-by-side with the CLI agents (claude, codex, gemini, copilot) in the same local-review review fan-out — no separate single-LLM-fallback path.

Removed in v0.15: the top-level provider: block (the v0 single-LLM-fallback mode). Loading a YAML file that still carries a provider: key now surfaces a migration error pointing at this shape. The migration is a verbatim field-rename — provider.base_url → llms.<your-name>.base_url, same for model / api_key_env. (Deprecated in v0.14, removed in v0.15; the CHANGELOG carries the full migration snippet.)

See examples/.local-review.yml for the full annotated example.

Default: all active agents (CLI + provider)

local-review review

Pick a specific model for one CLI agent

local-review review --claude-model claude-opus-4-7

Review (multi-agent: every authenticated CLI + every reachable provider endpoint runs in parallel)

local-review review [<base>] # canonical: current branch vs <base> (default: main) local-review staged # review git diff --cached (pre-commit) local-review commit [<rev>] # review one commit (default: HEAD) local-review branch [<base>] # alias of review for muscle-memory