Offline coding assistant with workspace retrieval, running Qwen2.5-Coder locally via llama.cpp.
OfflineCoder runs a local Qwen2.5-Coder model through llama-server (llama.cpp)
on 127.0.0.1. It makes no network calls — it works fully offline.
Your code and prompts never leave the machine.
Setup
This extension expects the model and runtime to be available locally. Download the
OfflineCoder bundle (model + runtime + all extensions) from the releases page:
Then either unzip it and run install.ps1, or install this extension's .vsix
directly. By default the model/runtime are read from ~/.offlinecoder
(configurable via the offlineCoder.* settings).
Commands
Open Chat — offlineCoder.openChat
Ask — offlineCoder.ask
Explain Selection — offlineCoder.explainSelection
Fix Selection — offlineCoder.fixSelection
Restart Local Model Server — offlineCoder.restartServer
Rebuild Workspace Index — offlineCoder.rebuildIndex
Show Server Status — offlineCoder.showStatus
Requirements
VS Code 1.85 or newer
Windows x64 (CPU-only build)
Privacy
The extension talks only to http://127.0.0.1:<port> and can be run with
llama-server --offline plus a Windows Firewall outbound block for a hard,
OS-enforced offline guarantee.
