AI code review that refuses to mark up your LLM bill

What it does Kodus reviews pull requests on GitHub, GitLab, Bitbucket, and Azure Repos using whichever LLM you supply—Claude, GPT-5, Gemini, Llama, or any OpenAI-compatible endpoint. You pay the model provider directly; the project takes no markup. It also tracks technical debt and delivery metrics, and adapts its feedback to your codebase over time via a “Kody” learning layer.

The interesting bit Kodus treats LLM access like a utility you pay directly rather than a bundled SaaS upsell. The AGPL-licensed core can be self-hosted entirely, with encrypted data handling and an explicit promise that source code is not used to train models.

Key highlights

• Bring-your-own-key for unlimited PR reviews on the free Community tier
• Plain-language custom review rules (“Kody Rules”) and context-aware memory
• CLI and CI/CD integration for local or pipeline-based reviews
• Monorepo architecture: separate apps/api, apps/webhooks, apps/worker, and apps/web services
• Anonymous daily heartbeat on self-hosted instances, opt-out via KODUS_TELEMETRY_DISABLED

Caveats

• Several premium features—SSO, RBAC, audit logs, and an engineering metrics cockpit—are reserved for paid Teams and Enterprise tiers
• The README’s feature comparison table lists some items as “Not listed on card” for paid tiers, which makes the exact Community/Teams boundary slightly unclear
• Enterprise tier oddly drops BYOK support, requiring use of a “Kodus AI Tokens API key” instead

Verdict Teams that want AI review without surrendering model choice or paying a SaaS markup on tokens should look here; developers wanting a fully managed AI coding suite with no API key management may find the BYOK setup extra work.