Kong-chat-bridge

Kong-chat-bridge lets VS Code route chat-model requests to model endpoints you control. It is designed for Kong API gateway deployments and other OpenAI-compatible, Anthropic, Gemini, or Ollama-style backends.

This extension is independent software. It is not affiliated with, endorsed by, or sponsored by Microsoft, GitHub, OpenAI, Anthropic, Google, or any upstream extension project.

Features

• Configure multiple model providers with separate API keys.
• Use OpenAI Chat Completions, OpenAI Responses, OpenAI Responses over WebSocket, Anthropic Messages, Gemini, and Ollama-style APIs.
• Store model-specific settings such as base URL, context length, output tokens, temperature, reasoning effort, and custom headers.
• Generate Git commit messages from the Source Control view.
• Show token usage and open configuration from the status bar.
• Import and export provider configuration.

Quick Start

1. Open the command palette.
2. Run Kong-chat-bridge: Open Configuration UI.
3. Add a provider with a base URL and API key.
4. Add one or more model entries.
5. Select a configured model from VS Code's chat model picker.

Example settings:

"kong-chat-bridge.baseUrl": "https://api.example.com/v1",
"kong-chat-bridge.models": [
	{
		"id": "gpt-5.5",
		"owned_by": "kong",
		"family": "gpt-5.5",
		"apiMode": "openai-responses-ws",
		"context_length": 256000,
		"max_completion_tokens": 8192
	}
]

Important Settings

• kong-chat-bridge.baseUrl: Global provider base URL.
• kong-chat-bridge.models: Model list shown through the VS Code model picker.
• kong-chat-bridge.retry: Retry policy for transient HTTP failures.
• kong-chat-bridge.delay: Delay between requests.
• kong-chat-bridge.commitLanguage: Output language for generated commit messages.
• kong-chat-bridge.readFileLines: Default line budget for file-reading tool results.
• kong-chat-bridge.copilotUtilitySmallFallbackModel: Optional explicit fallback model for Copilot's internal gpt-4o-mini utility-small family selection. Empty by default.

Model Family

Set family when the host should treat a custom endpoint like a known model family. For example:

{
	"id": "gpt-5.5",
	"family": "gpt-5.5"
}

If family is omitted, the extension tries to infer common model families from the model id and otherwise uses oai-compatible.

Copilot Utility-small Fallback

By default, Kong-chat-bridge does not advertise any compatibility alias for Copilot's internal gpt-4o-mini utility-small family selection. If a specific VS Code/Copilot build fails during tool-selection optimization and you explicitly accept routing that internal utility request to one of your configured models, choose a model in the Configuration UI or set:

"kong-chat-bridge.copilotUtilitySmallFallbackModel": "gpt-5.5"

Leave this setting empty to avoid exposing the compatibility alias. Enable it only as a network workaround, because the host may treat the fallback as a gpt-4o-mini family model and this can affect reasoning/thinking UI behavior.

Image Generation

For Responses-compatible gateways that expose OpenAI image-generation tools, add the image tool through the model's extra request parameters:

{
	"id": "gpt-5.5",
	"apiMode": "openai-responses-ws",
	"extra": {
		"tools": [
			{
				"type": "image_generation",
				"model": "gpt-image-2"
			}
		],
		"tool_choice": "auto"
	}
}

When the upstream returns generated image data, Kong-chat-bridge emits a VS Code image response part and also writes the image to ~/.kong/kong-chat-bridge/generated-images with a Markdown image link in the chat response.

Notes

• API keys are stored with VS Code SecretStorage.
• Provider-specific keys use kong-chat-bridge.apiKey.<provider>.
• The WebSocket Responses mode is intended for Kong API gateway-style upstreams that support per-run session reuse.