LiteLLM聊天

Requirements

• VS Code 1.108.0 or higher
• LiteLLM proxy running (self-hosted or cloud)
• LiteLLM API key (if required by your setup)

Quick Start

1. Install the extension from the VS Code Marketplace
2. Open VS Code's chat interface
3. Click the model picker → "Manage Models..." → "LiteLLM"
4. Add a server: enter a label, base URL (e.g., http://localhost:4000), and API key
5. Select models to add

Configuration

Custom Model Parameters (Optional)

Override default request parameters for specific models using the modelParameters setting. This is useful for models with specific requirements (like gpt-5 requiring temperature: 1) or to customize behavior per model.

To configure: Add to your settings.json:

{
  "litellm-vscode-chat.modelParameters": {
    "gpt-5": {
      "temperature": 1
    },
    "gpt-4": {
      "max_tokens": 8000,
      "temperature": 0.8,
      "top_p": 0.9
    },
    "claude-opus": {
      "max_tokens": 16000,
      "temperature": 0.5
    }
  }
}

Supported parameters: - max_tokens - Maximum tokens in response - temperature - Randomness (0.0-2.0) - top_p - Nucleus sampling (0.0-1.0) - frequency_penalty - Reduce repetition (-2.0 to 2.0) - presence_penalty - Encourage new topics (-2.0 to 2.0) - stop - Stop sequences (string or array) - response_format - Structured output / JSON mode - reasoning_effort - Thinking/reasoning control (for supported models) - seed - Deterministic output - And any other parameter supported by your LiteLLM and model provider backend

All modelParameters keys are passed through to LiteLLM — the extension does not filter or restrict which parameters you can set. The reserved key _replaceDefaults is extension metadata and is never forwarded.

Built-in defaults: The extension applies temperature: 0.7 by default. Some models (e.g., gpt-5.5) have built-in overrides that suppress the default temperature. User modelParameters entries merge on top of these defaults.

_replaceDefaults: Set "_replaceDefaults": true in a model entry to skip built-in request-parameter defaults for that model (for example, the default temperature) and use only the request parameters you supply from configuration:

{
  "litellm-vscode-chat.modelParameters": {
    "gpt-4": {
      "_replaceDefaults": true,
      "top_p": 0.9
    }
  }
}

Prefix matching: Configuration keys use longest prefix matching. For example, "gpt-4" will match "gpt-4-turbo:openai", "gpt-4:azure", etc. More specific keys take precedence.

Server-scoped parameters: In multi-server setups, prefix a key with the server label and / to scope parameters to a specific server. Server-scoped entries take priority over unscoped ones:

{
  "litellm-vscode-chat.modelParameters": {
    "gpt-4": {
      "temperature": 0.7
    },
    "Production/gpt-4": {
      "temperature": 0.3
    },
    "Dev/gpt-4": {
      "temperature": 0.9
    }
  }
}

Parameter precedence: Runtime options > User config > Defaults

Troubleshooting