AutoCode for Copilot

Local VS Code language model provider for AutoCode OpenAI-compatible proxy models.

• GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna
• Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5, and Claude Haiku 4.5

Need an AutoCode account? Register through this invitation link.

1. Run npm install.
2. Run npm run compile.
3. Press F5 from this folder to start an Extension Development Host, or package and install the VSIX.
4. In the Extension Development Host, run AutoCode: Set API Key. To use Claude models, also run AutoCode: Set Claude API Key.
5. Open Copilot Chat and pick an AutoCode model.

The provider registers vendor: "autocode" and exposes GPT-5.6 Sol, Terra, and Luna alongside the existing AutoCode models. It also exposes Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5, and Claude Haiku 4.5. Reasoning-capable models provide a Codex-compatible Thinking Effort setting: none, low, medium, high, xhigh, and max.

1. Run npm install.
2. Run npm run compile.
3. Run npm run package to generate a VSIX file for side loading.
4. Install the VSIX with code --install-extension autocode-copilot-provider-0.0.5.vsix or the VS Code Extensions UI.

1. Create or sign in to your Visual Studio Marketplace publisher account.
2. Confirm publisher in package.json matches your Marketplace publisher ID.
3. Create a Personal Access Token with the Marketplace Manage scope.
4. Run npx vsce login <publisher-id> and paste the PAT when prompted.
5. Run npx vsce publish 0.0.5.

If you want to verify the package before publishing, run npx vsce package --no-dependencies and inspect the generated VSIX.

• The default base URL is https://vip.auto-code.net; both default and Claude credentials use this same base URL.
• API keys entered via AutoCode: Set API Key and AutoCode: Set Claude API Key are stored separately in VS Code SecretStorage. Use the matching clear command to remove either key.
• Model IDs are prefixed with autocode- so they do not collide with built-in Copilot/Codex models.
• gpt-image-2 is not exposed here because this is a chat provider; image generation should use the /v1/images/* APIs.