Release Sieve: a local AI second look before you ship

Description and use

Release Sieve lets freelancers and small teams verify that commit changes match release notes using a local MiniLM‑L6‑v2 model without external APIs.

Problem

Freelance release maintainers and small product teams need a way to verify that commit changes are correctly reflected in release notes without relying on paid APIs or external services.

Outcome

Release Sieve enables local, privacy‑preserving semantic review of release notes, identifying missing or mismatched notes and providing a human‑reviewable report without external API calls.

Workflow

• Collect commit subjects, bodies, and filenames from a Git range or JSON file.
• Parse single‑release Markdown notes and validate bounded inputs.
• Encode messages and notes as 384‑dimensional vectors using MiniLM‑L6‑v2 via Transformers.js and ONNX Runtime.
• Rank note candidates by cosine similarity and apply a minimum score and runner‑up separation policy.
• Present review items with source evidence, similarity score, and suggested internal changes.
• Allow the user to edit notes, re‑run the model, and export the review snapshot locally.
• Optionally run the same workflow in a CLI for headless environments.

Evidence

Release Sieve takes a narrower approach. Paste commit subjects, load a JSON change file, and add the planned notes for one release.

Source

The first run downloads about 38 MB of model and WebAssembly assets from the same site. Your input text stays in the browser.

Source

Release Sieve compares commit evidence with the notes for one release.

Source

The browser uses single‑threaded WebAssembly in a worker; the CLI uses local CPU inference.

Source

The model’s weights and runtime are what make local semantic comparison possible.

Source