Is there AI code review that works offline?

DiffGuardian can review a pull request with AI running entirely offline, using a model server on your own machine instead of a cloud provider. Four local engines are supported — Ollama, LM Studio, llama.cpp's llama-server, and vLLM — each reachable only on that same machine, and offline review works on every plan, including Basic.

#Setting up a local model

1. Start a local model server: Ollama, LM Studio, llama.cpp's llama-server, or vLLM.
2. In Settings → AI providers, turn on that engine's row, then confirm or edit its address — loopback only, localhost or 127.0.0.1, port 1024 or higher.
3. Select Check now — DiffGuardian shows how many models it found.
4. Route review features to that engine in Settings → Model routing.

Each engine listens on a different default address (Ollama at http://localhost:11434, LM Studio at http://localhost:1234, llama.cpp at http://127.0.0.1:8080, vLLM at http://127.0.0.1:8000), and each row has its own optional API key field for a server started with its own key requirement. See Choosing models for full setup steps per engine.

#What "offline" covers

DiffGuardian only connects to a local model server running on the same machine — never a remote or self-hosted endpoint elsewhere on your network. Sign-in, billing, and plan status still go through DiffGuardian's servers over the network; it is the AI review calls, not the whole app, that stay local and offline.

#Limits

• Offline review needs a local model server actually running on your machine; DiffGuardian does not ship or install one for you.
• Quality and speed depend on the model and the hardware you run it on.
• Sign-in and billing still require a network connection to DiffGuardian's servers.