Ollama Copilot Bridge

Extension MVP for using Ollama Cloud models from VS Code chat through the official Language Model Chat Provider API.

This project is community-built and is not affiliated with GitHub, Microsoft, or Ollama.

What It Does

• Registers the ollama-bridge model provider in VS Code as Ollama Bridge.
• Connects to Ollama Cloud at https://ollama.com/v1 by default.
• Stores your Ollama Cloud API key securely with VS Code SecretStorage.
• Streams chat responses from Ollama Cloud into VS Code chat.
• Shows last-request context usage when hovering the Ollama Bridge status bar button.
• Can also be pointed at local Ollama by changing ollamaCopilot.baseUrl to http://localhost:11434.

Commands

• Ollama Copilot: Manage
• Ollama Copilot: Set API Key
• Ollama Copilot: Clear API Key
• Ollama Copilot: Refresh Models
• Ollama Copilot: Test Connection

Settings

{
  "ollamaCopilot.baseUrl": "https://ollama.com",
  "ollamaCopilot.openaiCompatiblePath": "/v1",
  "ollamaCopilot.defaultModel": "gpt-oss:20b",
  "ollamaCopilot.retryMaxAttempts": 4,
  "ollamaCopilot.retryBaseDelayMs": 1500
}

503 Server overloaded comes from Ollama Cloud. The extension retries temporary 429, 503, and 504 failures automatically before surfacing the error to VS Code.

Model Metadata

The bridge enriches models automatically with POST /api/show:

• Context size comes from *.context_length in model_info, with num_ctx fallback.
• Capabilities come from Ollama's capabilities array.
• Request multiplier is read if the provider returns one; otherwise it is estimated from general.parameter_count.

Context Usage Hover

Hover the Ollama Bridge status bar button after sending a chat request to see:

• input context used versus the model context window
• estimated response tokens
• total estimated tokens for the last request
• max output tokens
• request multiplier

This is a bridge-owned estimate based on the last request handled by the extension. The internal Copilot context indicator is controlled by VS Code/Copilot.

Agent Mode And Tool Calling

Ollama Copilot Bridge supports VS Code tool calling for compatible Ollama models:

• VS Code sends available Agent mode tools to the selected Ollama Bridge model.
• The bridge forwards tool definitions to Ollama's OpenAI-compatible chat endpoint.
• Streamed tool_calls are converted back into VS Code LanguageModelToolCallPart responses.
• Follow-up tool results from VS Code are sent back to Ollama as OpenAI-compatible tool messages.

Tool execution is still controlled by VS Code and GitHub Copilot. The model only requests a tool call; VS Code decides whether the tool is available, asks for confirmation when needed, runs the tool, and returns the result.

Development

npm install
npm run compile
npm test

Then press F5 in VS Code and run the extension in an Extension Development Host.

Set your Ollama Cloud API key from the Command Palette:

Ollama Copilot: Set API Key

Then run:

Ollama Copilot: Test Connection

Notes

This extension integrates with VS Code's official language model provider surface. It does not patch or replace the closed GitHub Copilot inline completion engine. In supported VS Code/Copilot Chat builds, contributed language models appear in the chat model picker.