vscode-unify-chat-provider

Bringing providers into Copilot Chat

Unify Chat Provider is a VS Code extension that integrates multiple LLM API providers into GitHub Copilot Chat using the Language Model API. The README lists what it adds: aggregation of free mainstream model channel configurations, one click setup with automatic syncing of official model lists, and migration from other tools. It supports the major API formats, namely OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Ollama Chat, and Gemini, and it claims deep adaptation to the special features of more than forty five providers with recommended parameters for over two hundred models. The headline convenience feature is one click use of your own account quotas for Claude Code, Gemini CLI, Antigravity, GitHub Copilot, OpenAI Codex, and xAI Grok, so you can route Copilot Chat through subscriptions you already pay for. The extension also provides code completion with several fill in the middle algorithms, commit message generation through the same UI entry point as VS Code, and full import and export of configuration through Base64, JSON, URL, or URI. The feature list emphasizes that it is out of the box and compatible, which is the main appeal for someone juggling several providers inside one editor. The source lives in the smallmain-vscode-unify-chat-provider repository on GitHub and is implemented in TypeScript, with the license recorded as MIT in the project metadata.

Configuration and migrations

The setup story has two tracks. One click configuration pulls a provider from a well known list, walks through authentication, shows a config import screen you can edit, and saves it so the models appear in Copilot Chat. Manual configuration is shown with DeepSeek as the example: you pick an API format such as OpenAI Responses, set the base URL, choose authentication such as an API key, and enable auto fetch of official models. The README is detailed about the screen flow and even notes where the in place documentation for each field lives. Migration from other apps and extensions is a first class feature: a command imports config from other applications, lists detected config file paths, and lets you import from a custom path or paste config content. Management covers unlimited provider configurations with multiple configs allowed for the same provider, as long as names stay unique. Model management allows unlimited models per provider and even supports a hash suffix like glm4.7#thinking to toggle thinking mode on and off, with the suffix stripped when the request is sent. Auto fetch periodically pulls the latest official model list and fills in recommended parameters from an internal database when the API does not return them. The settings also expose balance monitoring for a long list of providers using only the base URL and API key.

Code completion and commit messages

Beyond chat, the extension replaces VS Code's built in code completion with its own when active, and the README warns that once enabled it automatically disables the editor's native completion. It offers several algorithms: a Simple FIM implementation that works with any model, a Copilot replica of VS Code's core FIM and NES behavior, a Zed algorithm matching the Zed editor's edit prediction for Zeta models, an Inception algorithm for Mercury Edit 2, and a Mistral algorithm for Codestral. The docs recommend the Zed and Inception algorithms for better results. Commit message generation is exposed through dedicated commands and through a sparkle button in the Source Control view, and the README says it produces better messages than the default VS Code path through the same entry point. There is a caveat about VS Code background tasks consuming Copilot quota: some background work uses utility models by default, which can eat a free Copilot account, so the extension provides a quick setting to move those to other models. A second note says the extension uses experimental VS Code APIs and may prompt for administrator permission to enable them; without them, some models, commit messages, code completion, and presets are reduced or unavailable. Those tradeoffs are spelled out plainly so users know what they lose by declining the proposed APIs.

Editorial conclusion

The extension is written in TypeScript, requires VS Code 1.115 or later, and is published under the MIT license on the Marketplace.