Skip to content

Releases: darkmatter2222/GithubCopilotExit

Release list

GithubCopilotExit v2.0 — DGX Spark Support, Enhanced Documentation & Dynamic Proxy

GithubCopilotExit v2.0 — DGX Spark Support, Enhanced Documentation & Dynamic Proxy Latest
Latest

Choose a tag to compare

View all tags
@darkmatter2222 darkmatter2222 released this 28 Jun 01:00
v2
ed120ba

GithubCopilotExit v2.0

Major release with NVIDIA DGX Spark (GB10) support, comprehensive documentation overhaul, and enhanced hardware compatibility.

🎯 What's New

Dedicated DGX Spark Support:

  • Primary platform now optimized for NVIDIA DGX Spark (GB10) with 122 GB unified memory
  • Full hardware tier breakdown (Tier 1: 32GB+, Tier 2: 24GB, Tier 3: <24GB)
  • Clear VRAM requirement guidance for optimal performance

Enhanced Documentation:

  • Complete README rewrite with idiot-proof setup instructions
  • Visual documentation with dashboard screenshots and hardware comparisons
  • Hardware compatibility table showing exact VRAM requirements per GPU platform

New CLI Launcher:

  • copilot-dgx.bat — Specialized GitHub Copilot CLI launcher with optimized DGX Spark settings
  • Model selection (Qwen3.6-27B, Qwen3-Coder 27B, OBLITERATED variants)
  • Auto-configured for maximum performance

Infrastructure Improvements:

  • Dynamic FastAPI proxy auto-discovers models from Ollama (zero code changes needed)
  • MongoDB integration for persistent token tracking and analytics
  • Live dashboard with real-time TPS charts and request history
  • Containerized deployment support via Docker

✅ Key Features

  • Local AI — Zero API costs, complete privacy, no data leaves your machine
  • Multi-GPU Support — RTX 5090 (32GB), DGX Spark (122GB), RTX 4090/3090 (24GB)
  • Large Context — ~128K tokens on 32GB+ VRAM systems
  • Tool Calling — Full VS Code Copilot integration with file edits, terminal commands, search
  • Vision Support — Image understanding for debugging and diagrams
  • Thinking Mode — Deep reasoning chains for complex tasks
  • Streaming — Real-time token flow via SSE streaming
  • Analytics — Live dashboard at localhost:8001 with TPS monitoring

📋 Requirements

Component Minimum Recommended
VRAM 32GB+ (DGX Spark or RTX 5090)
OS Windows 10/11, Ubuntu 24.04 (DGX Spark)
Python 3.12+
Disk ~18 GB free for model download

🚀 Quick Start

git clone https://github.com/darkmatter2222/GithubCopilotExit.git
cd GithubCopilotExit
.\scripts\setup-local.ps1
.\scripts\start-proxy-local.ps1

For GitHub Copilot CLI users on DGX Spark:

copilot-dgx.bat

🔧 Hardware Support Matrix

- NVIDIA DGX Spark (GB10) — 122 GB unified memory, optimal performance
- NVIDIA RTX 5090 — 32 GB VRAM, full experience
- NVIDIA RTX 4090/3090 — 24 GB VRAM, limited to ~32K context tokens

📊 Architecture

VS Code Copilot / GitHub Copilot CLI
       │
       ▼
FastAPI Proxy (localhost:8001) → Dynamic model discovery
       │
       ▼
Ollama (localhost:11434) → Qwen3.6-27B or other models
       │
       ▼
NVIDIA GPU (CUDA/VRAM) → Local inference

MongoDB (optional, remote/analytics) — Token tracking & history

🎉 Why v2.0?

This release transforms GithubCopilotExit from an RTX 5090-focused project into a production-ready local LLM infrastructure stack with:

- Multi-platform GPU support
- Clear hardware requirements and VRAM guidance
- Professional-grade documentation
- Containerized deployment options
- Persistent analytics via MongoDB

Perfect for developers who want private, zero-cost AI coding assistants without cloud dependencies.
Assets 2
Loading

GithubCopilotExit v1.0 — Local Copilot-Style Coding with Qwen3.6, Ollama, and VS Code

GithubCopilotExit v1.0 — Local Copilot-Style Coding with Qwen3.6, Ollama, and VS Code

Choose a tag to compare

View all tags
@darkmatter2222 darkmatter2222 released this 16 Jun 13:35
v1
77e427b

GithubCopilotExit v1.0

GithubCopilotExit provides a private, local GitHub Copilot-style coding assistant for VS Code using Qwen3.6-27B, Ollama, and a FastAPI OpenAI-compatible proxy.

This release is focused on replacing cloud-dependent coding workflows with a self-hosted local stack that supports large-context coding, streaming responses, tool-calling workflows, and zero ongoing API cost.

Highlights

  • Local Copilot-style coding assistant for VS Code
  • Qwen3.6-27B model support through Ollama
  • FastAPI proxy with OpenAI-compatible endpoints
  • Large-context coding workflows
  • Streaming response support
  • Tool-calling support for agentic coding scenarios
  • Vision-capable local model workflow support
  • RTX 5090-class local GPU target
  • Private/offline-first design for sensitive codebases
  • No per-token API billing

Why This Exists

GitHub Copilot is useful, but cloud-hosted usage-based AI coding can become expensive, opaque, and difficult to control at scale.

GithubCopilotExit is an experiment in taking back control: run the model locally, keep code private, expose an OpenAI-compatible API, and connect it into VS Code workflows without relying on external model billing.

Included

  • FastAPI local proxy server
  • Ollama integration notes
  • VS Code configuration guidance
  • OpenAI-compatible request/response handling
  • Streaming support
  • Local coding assistant setup documentation

Target Setup

This project is designed for high-end local inference environments, especially RTX 5090-class systems running Ollama with Qwen3.6-27B or similar coding-capable local models.

Notes

This is an early release and should be treated as a practical local AI coding stack rather than a polished commercial replacement. Expect iteration around model compatibility, tool-calling behavior, prompt handling, context management, and VS Code integration.

Loading