Open source AI code review with full control over models and costs. Get your free trial now https://teapawt.net/t.co/lddfcRcE3V
Joined September 2018
- Tweets575
- Following573
- Followers573
- Likes401
You can now create your own code review agents in Kodus! 😎
We just launched Custom Review Agents (Beta), which work alongside Kody, each focused on areas like architecture, security, test coverage, or code duplication.
You can create an agent using one of our templates or import an existing skill from your repository.
Agents follow your project's documentation, can use tools like plugins, CLIs, and ast-grep, and run on demand for more focused PR reviews.
🔗 Learn more: docs.kodus.io/en/how_to_use/…
A lot of new stuff landed in Kodus this week.
The biggest update is the platform redesign: a new sidebar, cleaner settings, ⌘K search, and a clearer view of what happened in each Kody review.
I pulled together the main highlights below 👇
The self-hosted Doctor now checks review quality too
Besides validating the installation, Doctor now analyzes reviews from the last 7 days.
It shows:
• failed runs
• reviews that ran with reduced capabilities
• fallback model usage
• partially applied Kody Rules
• agents that hit their limits
• findings that couldn’t be posted
• 👍/👎 feedback
Thought that was it? 👀
Nope, we shipped even more last week.
Check out the full changelog here: docs.kodus.io/changelog
Kody from Kodus retweeted
Aconteceu uma coisa curiosa nessa sessão. Era a implementação de uma funcionalidade em que coloquei o Opus (high) para implementar e o Astra (xhigh) para fazer a revisão local.
Acontece que AMBOS deixaram passar 3 vulnerabilidades, e quem pegou mesmo foi a Kody, da @kodustech, usando GPT Terra (medium) durante o code review da PR.
Isso reforça que quem (ou o que) escreve o código não pode ser também o único responsável por revisá-lo. Ter um harness especializado em revisão de código é extremamente necessário.
A few new updates just landed in Kodus 👇
This update improves two important parts of the experience:
more control over which AI models your team uses for different scenarios, and a way to turn knowledge shared in PRs into reusable context for Kody.
Kody can now turn conversations into lasting context.
When you explain a team convention, clarify why a suggestion doesn’t apply, or share context worth keeping, Kody can recognize that the information may be useful in future reviews.
From there, she can suggest creating a memory, creating a Kody Rule, or updating an existing rule.
You stay in control: Kody proposes the action and asks for your confirmation before saving any changes.
And we shipped even more this past week 👀
Check out the full changelog here 👇
docs.kodus.io/changelog
Kody from Kodus retweeted
we’re working on a way to give Kody more context while reviewing a PR
in many teams, a frontend change depends on an API that lives in another repository. The same goes for shared libraries, contracts, and coding standards
with linked repositories, Kody can reference those repos while reviewing the current PR.
it only looks across repositories when the diff touches something that appears to cross a system boundary, such as DTOs, exports, schemas, or API routes. For internal changes, there’s no need to inspect the other repositories
findings still stay on the PR being reviewed. The linked repository simply provides additional context and evidence for the suggestion
the goal is to reduce cases where a change looks correct within one repository but ends up breaking something on the other side
what kinds of cross-repository dependencies tend to cause the most problems in your team’s reviews?
here are the latest Kodus updates 👇
one of the improvements in this release is to Vertex AI support for self-hosted installations: authentication can now use the identity already available in your cloud infrastructure, without relying on static keys
this reduces the need to store sensitive credentials and makes setup simpler
we also made improvements to our Jira, GitLab, Bitbucket, and Azure DevOps integrations
check out all the details here: docs.kodus.io/changelog
Kody from Kodus retweeted
Kody caught a bug in our own PR yesterday that would have silently revoked the bot seats customers pay for.
The same PR added the feature to assign seats to bots and a cron that revokes any seat not in the git member list.
Bots aren't in the git member list. Every file was correct, nobody on the team saw it, and it was going out Friday.
Dogfood your own product. This is the tweet.
Kody from Kodus retweeted
nov 2025, we stopped using AI as a copilot and started vibe coding.
PRs per dev went from 10 to 24 a month. Then we hit bugs the agents couldn't debug, often in code another agent wrote.
WTF does code quality mean when AI writes the code?
The changelog with last week’s Kodus updates is out ✨
One of the improvements is support for rules in .agents/rules/.
If your team keeps architecture standards, coding conventions, or other instructions in that directory, Kody can now find them and use them as part of the rules and review flow.
You can see more details on this update, along with the other improvements and fixes from last week, in the full changelog ↓
docs.kodus.io/changelog