Prelint

Prelint reviews pull requests against a team's written product specifications rather than against its code style. The problem it targets is specific to AI-assisted development: an agent can produce a change that compiles, passes tests and reads well in review, while quietly contradicting a pricing policy, a data-retention commitment or an architectural decision recorded somewhere the agent never looked.

It watches for six categories of drift. Business-logic changes that silently alter pricing or billing. Compliance violations around data retention and consent. Infrastructure redundancy, such as a new vendor added when an approved one already exists. Domain-language inconsistency, where the code stops using the team's own vocabulary. Scope creep. And strategic misalignment against decisions already recorded in the specs. It also reads the spec corpus itself, surfacing contradictions between specs, conflicting technology choices and duplicated effort.

A second surface lets non-engineers ask product questions grounded in the actual specifications, so a product manager or compliance officer can get an answer from the source of truth without interrupting an engineer mid-task.

Prelint integrates with GitHub and GitLab and reads specs in Markdown, YAML or other structured formats. Setup is installing the app and pointing it at the spec files; it activates on the next pull request with no new workflow to adopt. Pricing is usage-based at $1 per completed review with no per-seat charge, $10 in free credits to start, free access for open-source projects and a startup programme. On security, the company documents per-organisation tenant isolation, zero data retention, no model training on customer code, TLS 1.3 in transit and AES-256 at rest. Prelint, Inc. is a member of NVIDIA's Inception Program.