Overview

The Opilot extension brings the full Ollama model ecosystem into VS Code. After installation and a model pull, you can use local AI models in Copilot Chat, as inline code completions, and in a dedicated @ollama chat participant — all without leaving the editor.

Sidebar Layout

The Ollama activity bar icon opens a sidebar with four panels:

• Local Models — Models downloaded to your machine. Shows running state, VRAM usage, and capability badges.
• Cloud Models — Models streamed from Ollama Cloud (requires ollama login).
• Library — The full ollama.ai/library catalog for browsing and pulling.
• Modelfiles — Your custom .modelfile files for building personalized models.

Each panel has toolbar icons for filtering, grouping, collapsing, and refreshing. See Sidebar & Model Management for the full panel reference.

Copilot Chat Integration

Model Picker

Open Copilot Chat and click the model selector dropdown. All locally installed Ollama models appear under a 🦙 Ollama group. Select one to use it for the current conversation — no @ollama needed.

@ollama Chat Participant

Type @ollama at the start of any message to use the dedicated Ollama chat participant, which uses the currently-selected default model. The participant is sticky once activated in a thread.

@ollama refactor this function to use async/await

See @ollama Chat Participant for all the details.

Tool Calling

Compatible models (those flagged with the 🛠 tools badge) can invoke VS Code tools, MCP servers, and custom skills during a conversation:

@ollama what files in this project have the most test coverage gaps?

Tool calling follows the VS Code Language Model API tool loop: the model emits tool call requests, VS Code invokes the tool, and the result is fed back for the next response turn. If a model rejects tool schemas, the extension automatically retries the request without tools.

Vision

Models with the 👁 vision badge accept image attachments in Copilot Chat. Drag an image into the chat input or paste from clipboard. Images are automatically stripped for non-vision models to avoid prompt overflow.

Thinking Models

Models like DeepSeek-R1, Qwen QwQ, and Kimi that expose chain-of-thought reasoning display responses in two collapsible sections:

• Thinking — the model's internal reasoning steps
• Response — the final answer

The extension detects thinking models by model ID pattern and automatically applies the split-view format.

Privacy & Security

• Local model conversations are processed entirely on your machine.
• No usage data, telemetry, or conversation content is collected.
• Auth tokens for remote instances are stored in VS Code's encrypted secrets API.
• The extension communicates only with ollama.host (default: http://localhost:11434).