REDDIT · REDDIT// 4h agoTUTORIAL

Gemma 4 powers local VS Code Copilot

A LocalLLaMA user reports a working VS Code Insiders workflow using Gemma 4 26B-A4B Q8 through llama.cpp and the OAI Compatible Provider for Copilot extension. The setup keeps Copilot Chat's Ask, Plan, and Agent modes pointed at a local OpenAI-compatible endpoint, with reported generation around 60 tokens/sec on a Radeon AI PRO R9700.

// ANALYSIS

This is not a product launch, but it is a useful field report: local coding agents are getting close enough that the limiting factor is now workflow polish, context management, and tool-call reliability.

–OAI Compatible Provider for Copilot is the key bridge, letting developers reuse GitHub Copilot's VS Code UI with local or alternative model backends
–Gemma 4 26B-A4B looks viable for coding-agent experiments when paired with careful llama.cpp cache, context, and template settings
–The reported loops and tool-use failures are the real caveat: local inference speed does not automatically translate into reliable agent behavior
–For privacy-sensitive or cost-conscious developers, this pattern points toward a practical self-hosted Copilot alternative rather than just a novelty demo

// TAGS

gemma-4-26b-a4boai-compatible-provider-for-copilotllmai-codingideinferenceself-hostedgpu

DISCOVERED

4h ago

2026-04-23

PUBLISHED

4h ago

2026-04-23

RELEVANCE

7/ 10

AUTHOR

supracode