Info

Built to be
distrusted.

Veratex gives autonomous agents access to real code. We designed the system so no component can both run that code and merge it.

The architecture

Every job runs in a fresh sandbox that is destroyed when the job ends. The sandbox has limited access, and our servers check the result before anything can land in your repository.

Isolated sandboxes

We clone your code into a fresh micro-VM for each job, then destroy the VM when the job ends. Jobs never share an environment, so one customer's job cannot see another's.

Least-authority credentials

The sandbox gets a repository-scoped GitHub token that expires within an hour. Coding jobs may read and push code but cannot merge. Analysis and ideation jobs are read-only. The agent process never receives the token; only the surrounding clone and push steps can use it.

Server-side merge guards

Only our backend can merge, using a separate token. Before it does, it checks fork status, diff size limits, your automation settings, and test results. A sandbox cannot approve its own work.

Observed verification

Our orchestrator runs your test suites and records the result itself. It does not rely on the agent's report. A failing suite blocks the merge.

Authenticated callbacks

Each job reports back with a single-use secret that is checked in constant time. We claim each result once, so forged, replayed, or duplicate callbacks cannot publish or merge anything.

Prompt-injection containment

We treat issue text, pull requests, and repository files as untrusted input. They cannot grant the agent more authority. Veratex never pushes to or auto-merges a fork pull request because the fork controls both the code and its test signal.

Data protection

How we handle personal data is covered in the privacy policy.

Responsible disclosure

If you find a vulnerability in Veratex, email security@veratex.ai with enough detail to reproduce the issue. We ask that you:

We will acknowledge the report, keep you updated while we investigate, and will not pursue legal action for good-faith research that follows these guidelines.

Questions

For anything security-related that is not a vulnerability report, contact security@veratex.ai.