Local-first PR review for AI-built software. Bring GLM, Ollama, or any OpenAI-compatible endpoint — your keys, your machine, your rules.
-if (user.isAdmin = true) {+if (user.isAdmin === true) {Runs on your Mac, server, or runner. No hosted repo ingestion required. Your code stays where you choose.
Start with GLM/Z.ai, plug in Ollama or OpenAI-compatible endpoints, then bring Anthropic, OpenAI, Gemini, or your hosted model key as adapters mature.
Free for public open-source projects. Private repos, commercial usage, and auto-updates unlock through low-cost support tiers.
Every command runs on your box. Every review is local by default.
[01]$npm i -g neondiff[02]→added 1 package in 2.1s[03]✓installed neondiff public beta · exit 0[04]$neondiff init --provider=glm --model=glm-5.2[05]→→ probing provider … ok · api key found[06]→→ writing ~/.neondiff/config.toml[07]✓config ready · public repos free · exit 0[08]$neondiff review --repo owner/project --pr 42[09]→validating diff lines · redacting secrets · current head[10]→ND-01 critical src/auth/access.ts:2 assignment-in-condition[11]→ND-14 warning src/api/user.ts:88 unhandled-promise[12]→ND-22 info walkthrough update sticky summary[13]✓2 inline comments · 1 sticky walkthrough · exit 0$
The first pilot path because it is cheap, capable, and easy to run through ZCode-style review loops.
Point NeonDiff at a local model endpoint for fully local review experiments when your hardware can carry the run.
Bring OpenAI, Anthropic, Gemini, or any OpenAI-compatible provider key. NeonDiff never resells bundled tokens.
Supported companion context: agent-readable repo wikis through OpenWiki and code-intelligence packets through GitNexus.