🤗 OAI Compatible Provider for Copilot

Use frontier open LLMs like Qwen3 Coder, Kimi K2, DeepSeek V3.2, GLM 4.6 and more in VS Code with GitHub Copilot Chat powered by any OpenAI-compatible provider 🔥

✨ Features

• Supports almost all OpenAI-compatible providers, such as ModelScope, SiliconFlow, DeepSeek...
• Supports vision models.
• Offers additional configuration options for chat requests.
• Supports control model thinking and reasoning content show in chat interface.

  

• Supports configuring models from multiple providers simultaneously, automatically managing API keys without switch them repeatedly.
• Supports defining multiple configurations for the same model ID with different settings (e.g. thinking enable/disable for GLM-4.6).
• Support auto retry mechanism for handling api errors like [429, 500, 502, 503, 504].
• Support token usage count and set provider api keys in status bar.

  

Requirements

• VS Code 1.104.0 or higher.
• OpenAI-compatible provider API key.

⚡ Quick Start

1. Install the OAI Compatible Provider for Copilot extension here.
2. Open VS Code Settings and configure oaicopilot.baseUrl and oaicopilot.models.
3. Open Github Copilot Chat interface.
4. Click the model picker and select "Manage Models...".
5. Choose "OAI Compatible" provider.
6. Enter your API key — it will be saved locally.
7. Select the models you want to add to the model picker.

Settings Example

"oaicopilot.baseUrl": "https://api-inference.modelscope.cn/v1",
"oaicopilot.models": [
    {
        "id": "Qwen/Qwen3-Coder-480B-A35B-Instruct",
        "owned_by": "modelscope",
        "context_length": 256000,
        "max_tokens": 8192,
        "temperature": 0,
        "top_p": 1
    }
]

* Multi Api Mode

The extension supports three different API protocols to work with various model providers. You can specify which API mode to use for each model via the apiMode parameter.

Supported API Modes

1. openai (default) - OpenAI-compatible API

   • Endpoint: /chat/completions
   • Header: Authorization: Bearer <apiKey>
   • Use for: Most OpenAI-compatible providers (ModelScope, SiliconFlow, etc.)

2. ollama - Ollama native API

   • Endpoint: /api/chat
   • Header: Authorization: Bearer <apiKey> (or no header for local Ollama)
   • Use for: Local Ollama instances

3. anthropic - Anthropic Claude API

   • Endpoint: /v1/messages
   • Header: x-api-key: <apiKey>
   • Use for: Anthropic Claude models

Configuration Examples

Mixed configuration with multiple API modes:

"oaicopilot.models": [
    {
        "id": "GLM-4.6",
        "owned_by": "modelscope",
    },
    {
        "id": "llama3.2",
        "owned_by": "ollama",
        "baseUrl": "http://localhost:11434",
        "apiMode": "ollama"
    },
    {
        "id": "claude-3-5-sonnet-20241022",
        "owned_by": "anthropic",
        "baseUrl": "https://api.anthropic.com",
        "apiMode": "anthropic"
    }
]

Important Notes

• The apiMode parameter defaults to "openai" if not specified.
• When using ollama mode, you can omit the API key (ollama by default) or set it to any string.
• Each API mode uses different message conversion logic internally to match provider-specific formats (tools, images, thinking).

* Multi-Provider Guide

owned_by in model config is used for group apiKey. The storage key is oaicopilot.apiKey.${owned_by}.

1. Open VS Code Settings and configure oaicopilot.models.
2. Open command center ( Ctrl + Shift + P ), and search "OAICopilot: Set OAI Compatible Multi-Provider Apikey" to configure provider-specific API keys.
3. Open Github Copilot Chat interface.
4. Click the model picker and select "Manage Models...".
5. Choose "OAI Compatible" provider.
6. Select the models you want to add to the model picker.

Settings Example

"oaicopilot.baseUrl": "https://api-inference.modelscope.cn/v1",
"oaicopilot.models": [
    {
        "id": "Qwen/Qwen3-Coder-480B-A35B-Instruct",
        "owned_by": "modelscope",
        "context_length": 256000,
        "max_tokens": 8192,
        "temperature": 0,
        "top_p": 1
    },
    {
        "id": "qwen3-coder",
        "owned_by": "iflow",
        "baseUrl": "https://apis.iflow.cn/v1",
        "context_length": 256000,
        "max_tokens": 8192,
        "temperature": 0,
        "top_p": 1
    }
]

* Multi-config for the same model

You can define multiple configurations for the same model ID by using the configId field. This allows you to have the same base model with different settings for different use cases.

To use this feature:

1. Add the configId field to your model configuration
2. Each configuration with the same id must have a unique configId
3. The model will appear as separate entries in the VS Code model picker

Settings Example

"oaicopilot.models": [
    {
        "id": "glm-4.6",
        "configId": "thinking",
        "owned_by": "zai",
        "temperature": 0.7,
        "top_p": 1,
        "thinking": {
            "type": "enabled"
        }
    },
    {
        "id": "glm-4.6",
        "configId": "no-thinking",
        "owned_by": "zai",
        "temperature": 0,
        "top_p": 1,
        "thinking": {
            "type": "disabled"
        }
    }
]

In this example, you'll have three different configurations of the glm-4.6 model available in VS Code:

• glm-4.6::thinking - use GLM-4.6 with thinking
• glm-4.6::no-thinking - use GLM-4.6 without thinking

* Custom Headers

You can specify custom HTTP headers that will be sent with every request to a specific model's provider. This is useful for:

• API versioning headers
• Custom authentication headers (in addition to the standard Authorization header)
• Provider-specific headers required by certain APIs
• Request tracking or debugging headers

Custom Headers Example

"oaicopilot.models": [
    {
        "id": "custom-model",
        "owned_by": "provider",
        "baseUrl": "https://api.example.com/v1",
        "headers": {
            "X-API-Version": "2024-01",
            "X-Request-Source": "vscode-copilot",
            "Custom-Auth-Token": "additional-token-if-needed"
        }
    }
]

Important Notes:

• Custom headers are merged with default headers (Authorization, Content-Type, User-Agent)
• If a custom header conflicts with a default header, the custom header takes precedence
• Headers are applied on a per-model basis, allowing different headers for different providers
• Header values must be strings

* Custom Request body parameters

The extra field allows you to add arbitrary parameters to the API request body. This is useful for provider-specific features that aren't covered by the standard parameters.

How it works

• Parameters in extra are merged directly into the request body
• Works with all API modes (openai, ollama, anthropic)
• Values can be any valid JSON type (string, number, boolean, object, array)

Common use cases

• OpenAI-specific parameters: seed, logprobs, top_logprobs, suffix, presence_penalty (if not using standard parameter)
• Provider-specific features: Custom sampling methods, debugging flags
• Experimental parameters: Beta features from API providers

Configuration Example

"oaicopilot.models": [
    {
        "id": "custom-model",
        "owned_by": "openai",
        "extra": {
            "seed": 42,
            "logprobs": true,
            "top_logprobs": 5,
            "suffix": "###",
            "presence_penalty": 0.1
        }
    },
    {
        "id": "local-model",
        "owned_by": "ollama",
        "baseUrl": "http://localhost:11434",
        "apiMode": "ollama",
        "extra": {
            "keep_alive": "5m",
            "raw": true
        }
    },
    {
        "id": "claude-model",
        "owned_by": "anthropic",
        "baseUrl": "https://api.anthropic.com",
        "apiMode": "anthropic",
        "extra": {
            "service_tier": "standard_only"
        }
    }
]

Important Notes

• Parameters in extra are added after standard parameters
• If an extra parameter conflicts with a standard parameter, the extra value takes precedence
• Use this for provider-specific features only
• Standard parameters (temperature, top_p, etc.) should use their dedicated fields when possible
• API provider must support the parameters you specify

Model Parameters

All parameters support individual configuration for different models, providing highly flexible model tuning capabilities.

• id (required): Model identifier
• owned_by (required): Model provider
• displayName: Display name for the model that will be shown in the Copilot interface.
• configId: Configuration ID for this model. Allows defining the same model with different settings (e.g. 'glm-4.6::thinking', 'glm-4.6::no-thinking')
• family: Model family (e.g., 'gpt-4', 'claude-3', 'gemini'). Enables model-specific optimizations and behaviors. Defaults to 'oai-compatible' if not specified.
• baseUrl: Model-specific base URL. If not provided, the global oaicopilot.baseUrl will be used
• context_length: The context length supported by the model. Default value is 128000
• max_tokens: Maximum number of tokens to generate (range: [1, context_length]). Default value is 4096
• max_completion_tokens: Maximum number of tokens to generate (OpenAI new standard parameter)
• vision: Whether the model supports vision capabilities. Defaults to false
• temperature: Sampling temperature (range: [0, 2]). Lower values make the output more deterministic, higher values more creative. Default value is 0
• top_p: Top-p sampling value (range: (0, 1]). Default value is 1
• top_k: Top-k sampling value (range: [1, ∞)). Optional parameter
• min_p: Minimum probability threshold (range: [0, 1]). Optional parameter
• frequency_penalty: Frequency penalty (range: [-2, 2]). Optional parameter
• presence_penalty: Presence penalty (range: [-2, 2]). Optional parameter
• repetition_penalty: Repetition penalty (range: (0, 2]). Optional parameter
• enable_thinking: Enable model thinking and reasoning content display (for non-OpenRouter providers)
• thinking_budget: Maximum token count for thinking chain output. Optional parameter
• reasoning: OpenRouter reasoning configuration, includes the following options:
  • enabled: Enable reasoning functionality (if not specified, will be inferred from effort or max_tokens)
  • effort: Reasoning effort level (high, medium, low, minimal, auto)
  • exclude: Exclude reasoning tokens from the final response
  • max_tokens: Specific token limit for reasoning (Anthropic style, as an alternative to effort)
• thinking: Thinking configuration for Zai provider
  • type: Set to 'enabled' to enable thinking, 'disabled' to disable thinking
• reasoning_effort: Reasoning effort level (OpenAI reasoning configuration)
• headers: Custom HTTP headers to be sent with every request to this model's provider (e.g., {"X-API-Version": "v1", "X-Custom-Header": "value"}). These headers will be merged with the default headers (Authorization, Content-Type, User-Agent)
• extra: Extra request body parameters.
• include_reasoning_in_request: Whether to include reasoning_content in assistant messages sent to the API. Support deepseek-v3.2 or others.
• apiMode: API mode: 'openai' (Default) for API (/v1/chat/completions), 'ollama' for API (/api/chat), 'anthropic' for API (/v1/messages).

Thanks to

Thanks to all the people who contribute.

• Contributors
• Hugging Face Chat Extension
• VS Code Chat Provider API

Support & License

• Open issues: https://github.com/JohnnyZ93/oai-compatible-copilot/issues
• License: MIT License Copyright (c) 2025 Johnny Zhao