Diffsift is out. This is the first public build, and it’s what the download page installs.

• Review packs. Each pull request gets findings with severity, confidence, and evidence quoted from the actual diff, plus a risk level, a suggested review focus, and questions for the author. Packs are Markdown files on your disk.
• Review inbox. Turn on monitoring and Diffsift picks up new and updated pull requests while the app is open and reviews them by your rules: per-repository toggles, skip drafts, a size cap, and re-review on push. Finished packs land in the inbox with a Windows notification. Monitoring is off until you turn it on.
• Your vote at a glance. The inbox shows where you’ve already voted or reviewed, and merged pull requests are archived automatically. Abandoned ones stay visible.
• Azure DevOps and GitHub. Both are read-only. Diffsift never comments, approves, rejects, or merges. You copy the comments you agree with.
• Models. LM Studio, Ollama, and any OpenAI-compatible endpoint on your machine or network. If your policies allow it, OpenAI, Azure OpenAI (including custom-named reasoning deployments), Anthropic, and OpenRouter work too, with a confirmation before any code is sent.
• Multi-pass review for large pull requests, and review profiles for general, security, performance, test, and architecture review.
• Setup wizard that walks through the connection, model, and output folder, with token budgets you can adjust.
• Signed Windows installer. Per-user install, no admin rights needed. Licenses are verified offline, and the 14-day trial needs no account.

Comment posting and author opinion packs from the private preview are held back while they get more work.

Diffsift is in private preview while we finish launch packaging. What already works:

• Review packs — structured findings with severity, confidence, and evidence quoted from the actual diff, written as Markdown to your disk.
• Multi-pass review for large PRs: triage, summarise, deep-review the risky files, consolidate.
• Azure DevOps and GitHub pull request providers (GitHub is read-only for now).
• Review profiles — senior C#, security, performance, tests, architecture.
• Author opinion packs — an “is this ready?” pre-review for your own PR.
• Controlled comment posting to Azure DevOps: per-comment confirmation, duplicate protection, AI-assisted marker, and a local audit log.
• Local-first model support — LM Studio, Ollama, and OpenAI-compatible endpoints, with explicit warnings before any external provider is used.

Join the email list to hear when the public build ships.