Introducing Veritos: GitOps for AI Coding Agents
We run a dev agency in Vienna. Jumax. We do client work, internal tools, some open source stuff on the side. The usual. And like probably every dev team right now, Claude and Cursor have become a pretty big part of how we write code.
Here’s the thing nobody tells you about AI coding agents: the more repos you have, the worse it gets to keep the context files in sync.
How it started going wrong
We had .claude folders everywhere. Every repo had its own set of rules and skills. Some were up to date. Some were from months ago. A few had rules that one of us started writing and never finished.
Our Next.js coding standards lived in one repo. React Native patterns in another. And because we’d set them up at different times, they didn’t even agree on basics like naming conventions.
At some point we onboarded a client team onto one of our projects. They cloned the repo from a template, started working with Claude, and for days their AI was generating code with our old conventions. Nobody noticed until a PR review. The template they’d cloned from just had stale context and there was no mechanism to update it.
That was the moment we looked at each other and thought, okay, this is actually a problem.
What we wanted
Honestly, nothing complicated. We just wanted our AI rules and skills to work like code: stored in one place, version controlled, and pushed out to repos automatically. If you update a rule, every repo that uses it should get a PR with the change. The team reviews it, merges it, done.
Basically GitOps, but for AI agent configuration.
So we built it.
What Veritos does
Veritos (from Latin “veritas”, truth) is pretty straightforward. You get a central Library where your team creates and manages all your AI resources: rules, skills, subagents, MCP configs. Everything in one place.
One library instead of scattered folders. You write your React testing standards or organization conventions once. Every connected and allowed repo gets them. When you update something, you update it in one place.
Updates go through PRs. When you change a library item, Veritos opens pull requests in all the repos that use it. Your team reviews the changes the same way they review any code change. No more “wait, who updated that rule and when?”
Full version history. Every item is versioned. You can see who changed what, when, and roll back if something breaks. Nothing fancy, just proper version control.
Permissions that make sense for teams. Organization defaults, team overrides, project-specific settings, and repo-level tweaks. It cascades down the hierarchy the way you’d expect.
The whole thing connects to GitHub through a GitHub App. Files land in your .claude/ folders. There’s a lockfile that tracks what’s synced. It reuses existing branches and PRs when possible so you don’t get flooded with noise.
Why we built it ourselves
We’re not a VC-funded startup chasing a trend. We’re a software studio. We ship production code for clients every week. If a tool doesn’t hold up under real deadlines and messy codebases, it doesn’t survive long in our workflow.
We built Veritos because we needed it. Every feature exists because at some point someone on our team said “can we please fix this already.”
What’s on the roadmap
Right now we support Claude and Cursor (through Claude’s compatibility mode). More agent formats will follow as the ecosystem grows.
We’re also working on a Knowledge Base where you can upload your existing docs (Markdown, PDFs, whatever) and use AI to turn them into structured library items. So instead of rewriting your API documentation as Claude rules by hand, you feed it in and get usable context out.
A few other things we’re building:
- Veritos CLI so you can pull skills and rules directly into your local project via
npx veritos pull, with filters for team/project scope and a watch mode for automatic updates. - MCP Server that exposes your whole Veritos library as an MCP endpoint. Agents can discover and use approved skills at runtime without needing synced files in the repo.
- Open Agent Skills Standard support, so skills can be full directories with a
SKILL.mdplus subfolders likescripts/,references/,assets/, and whatever else you need. - Unified granular permissions across users, teams, projects, and all resource types. Proper role-based access control with hierarchical inheritance.
There’s more on the full roadmap, including project import, project templates, automatic repo indexing, and connectors for external knowledge sources like Confluence and Notion.
We’re shipping now because the core problem (keeping AI context in sync across repos) is solved. Everything else builds on top of that.
Try it out
If you’re juggling .claude folders across multiple repos, or you’re starting to notice that your AI gives different advice depending on which project you’re in, give Veritos a look. It’s free to start with.
Built by Julian Pomper and Max Gierlachowski in Vienna.