Introduction

CodeOtter is a free-to-self-host pull request review platform for engineering teams who want calibrated code intelligence without sending private code to a third-party SaaS or paying per seat.

Every review produces:

• 6 calibrated 0–100 gauges: Quality, Blast Radius, Risk, Tests, Readability and PR Hygiene.
• 9 binary merge gates: Title, Description, Security, Complexity, Tests, Docs, Scope, Guidelines and Issue Requirements.
• A written review: walkthrough, file cohorts and severity-ranked findings with committable suggestion fixes.

Run it 100% offline with open-weight Hugging Face models (llama.cpp and ggmlc sidecars), or connect your own key for Claude, OpenAI, MiniMax or OpenRouter.

Start hereCopy Anchor Link

CodeOtter is right for your team ifCopy Anchor Link

• You want PR reviews without sending private code to a third-party cloud.
• You want calibrated 0–100 scores and deterministic gates, not hallucinated guesses.
• You want to enforce team conventions (AGENTS.md / CLAUDE.md) automatically on every PR.
• You want to run 100% offline on local hardware (MacBook, Linux workstation, air-gapped GPU).
• You want zero per-seat billing: self-hosting is free.
• You want a small deployment: one Docker container or a Node process.

Supported models and providersCopy Anchor Link

Mix and match System 1 and System 2 engines freely:

• Hosted (BYOK): Claude Opus / Sonnet, OpenAI GPT-5, MiniMax-M3, OpenRouter, Cloudflare Clef, TypeSafe Jev.
• Local / open weights: Qwen2.5-Coder (1.5B / 7B), DeepSeek-Coder / V3, Llama 3.1 / 3.3, Microsoft CodeReviewer, Laya Typed-Decisions, Kev 0.8B / 4B, Ollama.
• Anything else that speaks the OpenAI API or ships as a GGUF.

LicenseCopy Anchor Link

CodeOtter is open source under the GNU AGPL-3.0. It is free to self-host, modify and redistribute; anyone who runs a modified version as a network service must publish their changes.