Open source AI code review you can self-host, with the LLM you choose.
Kodus reviews pull requests on GitHub, GitLab, Bitbucket, Azure DevOps, and Forgejo. Run it on your own servers or use Kodus Cloud.
Website · Community · Docs · CLI Docs · Try Kodus Cloud · Self-Host Guide
🌐 English · Português (BR) · Español · 日本語 · 简体中文 · Français
- Model Agnostic: Use Claude, GPT-5, Gemini, Llama, GLM, Kimi or any OpenAI-compatible endpoint.
- Zero Markup on LLM Costs: You pay model providers directly. No hidden multipliers.
- Learns from Your Context: Kody adapts to your architecture, standards, and workflow.
- You Set the Rules: Define custom review rules in plain language.
- Privacy & Security: Source code is not used to train models, data is encrypted in transit and at rest, and self-hosted runners are supported. Self-hosted instances send one anonymous heartbeat per day. It carries aggregated counters and no code, names, or identifiers. Opt out with
KODUS_TELEMETRY_DISABLED=true. See Anonymous Telemetry. - Native Git Workflow: Works directly in PRs with GitHub, GitLab, Bitbucket, Azure DevOps, and Forgejo.
- CLI + CI/CD Ready: Run reviews locally and in pipelines.
- Operational Impact: Track technical debt and delivery metrics while keeping review quality high.
🔑 Bring Your Own Key
Connect your own provider credentials and choose the models behind Kodus reviews: OpenAI, Anthropic, Google Gemini, Vertex AI, Novita, or any OpenAI-compatible endpoint. Keep billing and usage under your own provider account, without hidden LLM markups.
📈 Token Usage
Track token consumption across AI code reviews, understand cost drivers, and keep model spend predictable as adoption grows.
⚙️ Kody Rules
Kody Rules let teams define review instructions in plain language and apply them across organizations, repositories, paths, or specific review scopes. Kody uses those rules as context when reviewing pull requests, helping enforce architecture decisions, security expectations, testing practices, and repository-specific conventions without relying on reviewers to repeat the same feedback manually.