Kodus

Review bot where you control the model

Kodus reviews pull requests with an agent named Kody. The distinguishing choice is that it is model-agnostic and open source: the code is on GitHub under AGPLv3 (with a separate enterprise licence file for some features), and you point it at your own OpenAI, Anthropic or Gemini key. Its pricing page calls it the open source alternative to CodeRabbit. It works with GitHub, GitLab, Bitbucket, Azure Repos and Forgejo, which is a wider list than many hosted competitors, and self-managed GitLab and Forgejo are called out specifically.

Bring-your-own-key billing

You pay Kodus for the platform and your model provider directly for tokens, with no markup stated. That makes the cost visible in a way flat-fee review tools hide it, and it lets you choose a cheap model for routine diffs. The vendor's own calculator for a 30-developer Teams plan estimates about $345 a month using Gemini Flash and about $570 using Sonnet 4.5. Those are estimates; a monorepo with large diffs will cost more, and a quiet week costs less.

Rules, not just comments

Kody Rules let you write team conventions in plain language, such as "service handlers must not call the database directly" or "new endpoints need an auth decorator", and apply them per repository. Kody also keeps learnings and memory from past reviews, so repeated dismissals should reduce repeated noise. Whether that works well depends on how consistent your reviewers are about replying to its comments; there is no substitute for feedback on the bot's output.

Plans and limits

• Community: free, self-hosted or hosted, unlimited PRs and users with your own key. Capped at 10 Kody Rules and 3 active plugins; community support is on Discord.
• Teams: $10 per developer per month, or $8 annual, plus tokens. Hosted by Kodus only. Includes a 14-day trial limited to 35 PR reviews, unlimited rules and plugins, a priority queue, custom agents and an engineering metrics view.
• Enterprise: custom. Adds SSO, RBAC, audit logs, SOC 2 and up to 5 hours a month of dedicated support.

A pilot programme offers 30 to 100 percent discounts to startups, open source projects and teams in countries with weaker currencies.

The cost of running it yourself

The self-hosted deployment is Docker on a VM you manage, with at least 2 CPU cores, 8 GB RAM and 60 GB of free disk. Self-hosted instances send one anonymous heartbeat a day unless you set KODUS_TELEMETRY_DISABLED=true. Note the split: self-hosting is available on Community and Enterprise, not Teams, so a team that outgrows ten rules and wants self-hosting has to talk to sales.

Trade-offs

The GitHub repository shows an empty Releases section, so version pinning for a self-hosted install means tracking commits or tags yourself. Review quality depends on the model you choose, which is both the point and the risk.