How it works

Inside Diffnix
PRInspector Engine

PRInspector analyzes pull requests for security risks, architectural violations, and code-quality issues, helping engineering teams apply consistent review standards across their codebases. Here's how a review works. PRInspector is in pre-launch development.

Workflow Phase 01

Pull Request Webhook

When a developer opens or updates a pull request, the Diffnix GitHub App receives the webhook, verifies its signature and queues a review job with the PR metadata (repository, PR number, commit, author).

System Data Flow
INPUT:VCS Event webhook payload
OUTPUT:Scheduled task in queue

01 / Review Context

PRInspector reviews the files changed in a pull request together with the PR title and the rules enabled for the repository, so findings are specific to the change being made. Deeper cross-file analysis, which pulls in related modules the change depends on, is planned.

02 / Rules and AI Together

Deterministic rules catch well-defined problems, while AI analysis handles issues that need context — for example, telling a real hardcoded secret apart from code that reads one from the environment. Large pull requests switch to a summary review so feedback stays useful.

03 / Inference You Control

PRInspector supports self-hosted inference with open models served through Ollama, for customers requiring private deployments. Optional cloud AI providers are planned for teams that choose hosted inference. Your inference configuration determines how code is processed.

PRInspector Is in Pre-launch Development

We're building AI-powered code review tools for engineering teams. Join the waitlist to follow development and hear when early access becomes available.