Opening project dossier…
Opening project dossier…
A local-first code impact explorer that maps file connections from Git evidence and reviews changes with cited AI.
A diff shows which lines changed, but not who imports the file, which tests usually change with it, or what else is at risk.
AI explanations are hard to trust when their evidence is hidden. Atlas keeps repository facts and model interpretation visibly separate, and shows failures instead of smoothing them over.
Making AI review auditable: every claim has to point at a line in the diff, and an incomplete run has to look incomplete.
A React and TypeScript interface talks to a local Express API that reads imports and bounded Git history without executing repository code. Review runs on local Ollama, or on Ollama Cloud only after a consent preview. Model replies are checked against the changed lines, and feedback lives in local SQLite.
Import detection is pattern-based rather than a full parser: faster and language-agnostic, at the cost of edge cases.
Diffs are capped at 38,000 characters, and repositories using Git LFS or clean filters are not supported for previews.
Agreement between review passes is reported as agreement, not as a calibrated probability.
Synthetic walkthrough
Status source: Public site is a synthetic walkthrough; the README gates a release on the first passing remote CI run.
Project references: stxic-atlas README · GitHub repo description · christianlascuna.dev
Let’s talk about your web, mobile or desktop app, automation, AI, QA, or CRM project.
Discuss your project