9Router for GitHub Copilot
Use 9Router and any OpenAI-compatible model inside GitHub Copilot Chat — with automatic fixes for the rough edges that self-hosted models produce.
Quick Start
- Install from the VS Code Marketplace.
- Open Settings → search
9Router → set Server URL to your endpoint (e.g. http://localhost:20128/v1).
- Use the Configure Server command (
Cmd+Shift+P → "9Router: Configure Server") to set your API key.
- In Copilot Chat, open the model picker → Manage Models → enable models under 9Router.
- Select a model and start chatting.
Why This Over Native BYOK?
VS Code's built-in BYOK works great for well-behaved models. 9Router adds a resilience layer for when things break:
- Tool calls fail with bad JSON? Repairs truncated arguments and fills missing required fields.
- Reasoning models leak
<think> blocks? Routes them into Copilot's thinking UI, not your chat.
- Context-length errors? Auto-detects the real limit and budgets tokens safely.
- Model outputs tool names as text? Lowers temperature and stabilises formatting.
Inference stays on your server. No per-token fees. Doesn't consume Copilot premium quota.
Servers Supported
9Router, vLLM, Ollama, llama.cpp, LM Studio, LocalAI, LiteLLM — any OpenAI-compatible endpoint.
Key Settings
| Setting |
Default |
What it does |
9router-for-github-copilot.serverUrl |
http://localhost:8000/v1 |
Your inference server base URL |
9router-for-github-copilot.defaultMaxTokens |
262144 |
Fallback context window |
9router-for-github-copilot.enableToolCalling |
true |
Allow agent tools (file ops, terminal, etc.) |
9router-for-github-copilot.agentTemperature |
0 |
Tool-call stability (lower = stricter) |
9router-for-github-copilot.modelContextWindows |
{} |
Per-model context overrides, e.g. {"qwen3-8b": 32768} |
9router-for-github-copilot.perModelOptions |
{} |
Per-model sampler params (temperature, top_p, etc.) |
Commands
| Command |
Purpose |
| 9Router: Configure Server |
Set server URL and API key |
| 9Router: Test Server Connection |
Verify connectivity and list models |
| 9Router: Refresh Models |
Re-probe the server for model changes |
| 9Router: Edit Custom Headers |
Manage custom HTTP headers (stored securely) |
| 9Router: Show Output Log |
View debug output |
Troubleshooting
Models not showing? Run curl <server-url>/v1/models to verify the server is up. Make sure the URL has no trailing /v1 or slash. Run the Test Server Connection command for a diagnostic.
Tool calls output as text instead of executing? Set Agent Temperature to 0, disable Parallel Tool Calling, and confirm your server has --enable-auto-tool-choice (vLLM).
Context overflow errors? Add the model to modelContextWindows with the correct limit. The extension learns the real limit from the error and retries once automatically.
Models not in the Agents window? Agents window runs in a separate process. Add this to settings and reload:
"extensions.supportAgentsWindow": {
"hotrungnhan.9router-for-github-copilot": true
}
Utility Tasks (Titles, Commit Messages)
By default, Copilot sends chat titles and commit messages to GitHub. Route them to your own model instead:
- Open Settings → set
chat.utilityModel and chat.utilitySmallModel to a 9Router model.
Privacy
Prompts and code go to your server only. Copilot Chat (the VS Code host) may send its own auth, telemetry, and title requests to GitHub — route utility tasks to your server to minimise this.
Credits
This project is a fork of arbs-io/github-copilot-llm-gateway. Thanks to the original authors for the base extension.
License
MIT — see LICENSE.
