No GitHub sign-in. BYOK models for Copilot Chat & Agent Mode — just an OpenAI-compatible URL and key.
Quick start
Install and enable the extension, then Reload Window (requires VS Code 1.109+ with GitHub Copilot Chat installed).
Run Custom Mode: Manage Models… from the Command Palette (or click the ✨ status bar icon). On first use you go straight into the add-model wizard:
① API Base URL (OpenAI-compatible, e.g. https://api.example.com/v1)
② Display alias (the name shown in the model picker; blank = show the model ID)
③ Real model ID (the model parameter sent to the endpoint)
④ API Key (leave blank for local/intranet gateways without auth)
Select it in the Copilot Chat model picker (BYOK group) and start chatting.
Model groups are stored in the user-level credential store (not settings.json): shared across profiles and windows, applied immediately, no reload required.
Commands
Custom Mode: Manage Models — one dialog to add / update model ID / duplicate / delete model groups.
Custom Mode: Configure Agents Window Tools — bring tools to the front in the Agents window (reversible).
Custom Mode: Reset Learned Reasoning Effort Limits — clear the record of which reasoning levels each endpoint rejected.
Troubleshooting
Model doesn't appear in the picker: make sure Copilot Chat is enabled.
Gateway returns 400: the extension falls back automatically through a four-level ladder; see the Output panel (Output → "Custom Mode").
"Request too long": lower "Max input tokens" in the gear menu next to the model, or start a new session.
Post-hoc debugging: request logs are persisted to custom-for-copilot.log in the extension's global storage (macOS: ~/Library/Application Support/Code/User/globalStorage/CustomforCopilot.custom-for-copilot/).
Development
npm install && npm run compile
npm run verify # compile + i18n check + smoke tests
./scripts/dev-install.sh # package and install into local VS Code (Reload Window after)
