🔥 Testing Devstral with Ollama and GitHub Copilot in VS Code

🎯 Objective

Evaluate the performance and integration of the Devstral model from Mistral AI, running locally using Ollama, within the developer environment of Visual Studio Code (VS Code) using GitHub Copilot.

🧩 Components Used

• Ollama: Local model runner (https://ollama.com)

• Devstral: New coding-focused model from Mistral (Devstral info)

• Visual Studio Code: Code editor (https://code.visualstudio.com)

• GitHub Copilot: AI assistant by GitHub (https://github.com/features/copilot)

🛠️ Setup Instructions

Step 1: Install Ollama

To run Devstral locally:

1. Go to https://ollama.com

2. Download and install Ollama for your OS (Windows/Mac/Linux).

3. Once installed, open a terminal and run:

ollama run devstral

This will download and prepare the Devstral model for local inference.

Step 2: Install Visual Studio Code

1. Visit https://code.visualstudio.com

2. Download the latest version for your OS and install it.

3. Install recommended extensions:

   • Python (if you’re working on Python projects)

   • GitHub Copilot

Step 3: Activate GitHub Copilot

1. In VS Code, go to Extensions (Ctrl+Shift+X) and install GitHub Copilot.

2. Sign in to GitHub when prompted.

3. Copilot should now be active in your editor. You’ll start seeing autocomplete suggestions as you type.

Step 4: Enable Ollama + Devstral in VS Code

Github Copilo offer now the option to set Ollama with you local models:

As shown in the screenshot you’ll now see Ollama: Devstral listed as an available model alongside Claude, GPT, etc.

When you clic on Manage Models you can select your models

When you click on Ollama, you can set the locam models that you have installed.

⚙️ Observations

• Devstral performed quite well in code generation, particularly for Python and web frameworks .

• It's fast and responsive when run locally via Ollama, and it's a solid alternative to cloud models, the speed depends of your GPU.

• However, the Agent Mode in Visual Studio Code didn’t work correctly during tests — the agent mode worked as the Ask mode.. This might still be an early integration issue or a compatibility bug in VS Code plugins.

✅ Conclusion

Running Devstral locally using Ollama and integrating it into VS Code offers a promising experience for developers who want fast, local LLM performance with reduced cloud dependency. It's particularly useful when working on private codebases or when you need consistent latency.

That said, Agent Mode support is not stable yet, so it’s best used in autocomplete/code-completion contexts for now. As the ecosystem matures, we expect full support to arrive soon.