What it does
Four things, and two of them you never have to think about.
Two products you use directly, and two layers underneath that make them possible. Nothing here is a roadmap item — this is what runs today.
- A symbol graph of your repository — files, functions, classes, and the imports, calls and inheritance between them.
- Sixteen agents that produce findings: nine read your architecture, seven watch dependencies, secrets, SAST and infrastructure config.
- One index out of 100, across four measures, with every deduction traceable to a file path.
- Every finding ships with the prompt that fixes it.
- Runs on every pull request through a webhook, plus a nightly pass.
- Ask your codebase why — grounded in the graph, not a guess.
- Live presence — who is in the codebase right now, teammates and their agents alike.
- Declared intents: an agent says what it is about to touch, before it touches it.
- Overlap checked against everyone in flight, including one hop of reverse dependencies.
- Collisions come back as proposals in context, never as failed calls.
- A landing queue that rebases, gates, pushes and opens the pull request.
- Publish a skill once and every agent on your team has it.
Remember what outlives a single scan.
A layer, not a product — there is nothing here to sign up for. It is why the two above work the way they do.
- The shape of your repository at any commit you ask for.
- Search your code by what it does, not what it was named.
- A record that keeps both timelines — what was true, and when you came to know it.
- New snapshots add to history rather than overwrite it.
Your rules on your own model usage — if you want them.
A layer, not a product — there is nothing here to sign up for. It is why the two above work the way they do.
- Four enforced policy types: model access, rate limits, content filtering and data routing.
- One switch stops it — a model, a team, a vendor, or everything.
- Every call costed and logged, with budgets that warn on the way up.
- Role-based access including a read-only seat for whoever asks the hard questions.
- An audit trail the database itself refuses to let you edit.
How the pieces fit
Two products that share one memory.
And a control plane for all the AI your team uses — sitting somewhere else entirely, and only if you want it.
Gradus knows two agents are about to collide because VibeCop read the code and Partikle kept the graph. That is not an integration — it is the same system.
See it on a real repository.
There is a full report on this site, run against a repository we built to exercise every detector. No signup, no repository, no waiting.