A Visual Studio 2026 extension that turns any OpenAI-compatible model — local or remote — into a code-aware, read-only assistant scoped strictly to your open solution.
Point it at your own endpoint (Ollama, LM Studio, vLLM, or the OpenAI API). It reads only the currently-open solution, shows you exactly where your code is being sent, and suggests changes in chat for you to apply by hand. It never writes to your files.
Most in-IDE AI assistants assume a specific cloud provider and send your code there. This one assumes the least: any model behind a URL, running wherever you want — including entirely on your machine. It's for people who want their own model, their own data boundary, and a tool that's honest about both.
- Solution-only. The model can read only files that belong to the currently-open solution. With no solution open, it's disabled. Files outside the solution — even ones open in another editor tab — are refused.
- You choose the endpoint. The only place your code is sent is the API URL you configure. The extension does not phone home.
- You can see where it goes. A chip under the input shows the destination and whether the connection is encrypted (🟢 this machine · 🟡 external + https · 🔴 external + http), and asks once before anything first leaves your machine.
- Read-only. It suggests edits in chat; you select, copy, and apply them yourself. It does not modify your files.
- Agentic, code-aware chat. By default every message runs a tool-using loop that inspects your actual code (Roslyn find-references / symbols / outlines, content search, file read) instead of guessing.
- Any OpenAI-compatible model. Native
tool_callswhen the endpoint supports them; a text-based tool protocol fallback for models that don't — so even a modest local model works. - Streaming responses with a collapsible "thinking" view.
- Editor right-click actions: Explain, Refactor, Generate Tests, Add Doc Comments (limited to in-solution files).
- Prompt history in the input box (Ctrl+↑ / Ctrl+↓).
- Capability probe / Test Connection so you know what your endpoint supports.
- Context-budget management to keep long sessions inside your model's window.
- Visual Studio 2026 (version 18).
- An OpenAI-compatible chat-completions endpoint (e.g. Ollama, LM Studio, vLLM, or
api.openai.com).
From the Visual Studio Marketplace (search "Local Model Integrator"), or download the .vsix from the Releases page and double-click it.
- Tools → Options → Local Model Integrator.
- Set API URL, Model Name, and Protocol (OpenAI Compatible / Ollama / DeepSeek vLLM / LM Studio).
- Click Test Connection to confirm reachability and capabilities.
- Open the chat from View → Other Windows → Local Model Integrator, open a solution, and ask away.
Useful options: Context Window (tokens), Agent Max Steps (0 = unlimited), Request Timeout (ms) (0 = no timeout), Agentic Chat (toggle tool use), Enable Reasoning / Streaming.
Requires VS 2022 with the Visual Studio extension development workload.
msbuild src\LocalModelIntegrator\LocalModelIntegrator.csproj /t:Restore
msbuild src\LocalModelIntegrator\LocalModelIntegrator.csproj /p:Configuration=ReleaseThe VSIX lands in src\LocalModelIntegrator\bin\Release\net472\.