This extension allows any BYOK (Bring Your Own Key) AI editor or extension, such as Cursor or Continue, to connect to any HTTP-compatible LLM by aliasing it as a different model (e.g., gpt-4o). The extension starts a local server that proxies requests, enabling you to bypass restrictions and unlock features that are often limited to specific models like Claude or OpenAI's GPT series.
Many AI-powered IDEs and extensions that support BYOK still restrict access to advanced features—such as Agent mode, image and voice inputs, or multi-modal capabilities—to a small, hardcoded list of models. This means that even if you have access to a powerful open-source model, you might not be able to use it to its full potential.
LLM Proxy solves this problem by acting as a middleman. It intercepts requests from your editor, swaps the model alias with the real model name you want to use, and forwards the request to your self-hosted or cloud-hosted LLM. This tricks the editor into thinking you are using a supported model, unlocking all its features for use with any compatible LLM.
- Universal Compatibility: Works with any HTTP-compatible LLM, including Ollama, OpenRouter, and custom endpoints.
- Unlock Restricted Features: Enables Agent mode, multi-modal inputs, and other advanced features in popular AI extensions.
- Simple Configuration: Easily add and manage model aliases through commands.
- Automatic Start-up: The proxy server starts automatically when you open.
- Status Bar Indicator: Quickly check the proxy's status from the status bar.
- Install the LLM Proxy extension from the Marketplace.
- Open the command palette (
Ctrl+Shift+PorCmd+Shift+P). - Run the LLM Proxy: Add Model command.
- Follow the prompts to configure your model:
- Alias: Enter a well-known model name to trick the editor (e.g.,
gpt-4o,claude-3-opus-20240229). - URL: Provide the base URL of your LLM endpoint (e.g.,
http://localhost:11434for Ollama). - Real Model: Enter the actual model name you want to use (e.g.,
llama3,codegemma).
- Alias: Enter a well-known model name to trick the editor (e.g.,
Once the proxy is running, configure your AI extension to use the local proxy server:
- API Base URL:
http://localhost:4891/v1 - Model: Select the alias you created (e.g.,
gpt-4o).
That's it! Your editor will now send requests to the local proxy, which will forward them to your chosen LLM while unlocking all its features.
This extension is licensed under the MIT License. See the LICENSE file for details.