Video summary

One-Click Local AI in VS Code with Language Model Providers

Main summary

Key takeaways

Technology

Summary of Key Technological Updates & Features (VS Code + Local AI)

New VS Code Updates for Configuring Language Model Providers

VS Code introduces improvements to make it easier to add and configure multiple LLM providers within the editor, including support for:

  • Local models running on your machine
  • “Bring your own key” (BYOK) providers (you supply API keys)
  • Custom endpoints

One-Click Installation of the Ollama Extension

A new extension—specifically highlighted: the Ollama extension—can be installed with one click.

After installation, it:

  • Automatically connects VS Code chat/agents to Ollama models running locally
  • Includes Ollama-specific commands such as:
    • Refresh
    • Diagnosing models (to troubleshoot issues)

Model Selection and Provider Setup

In VS Code, users manage this through model settings, where they can configure language model provider settings and add API keys for many providers, including:

  • Anthropic, Azure, Google, OpenAI, OpenRouter, xAI
  • Custom endpoints

Users can also install additional providers from the VS Code Marketplace, such as:

  • Foundry, DeepSeek, Hugging Face
  • The Ollama extension (highlighted as “brand new”)

Utility Model Customization

VS Code adds utility model settings for small automated actions/flows, including:

  • Commit messages
  • Renaming session titles

Users can choose utility models from:

  • Configured BYOK providers (e.g., OpenRouter via an API key)
  • Local Ollama models

A BYOK utility model mode is mentioned that can use:

  • Defaults
  • Main agent
  • GitHub Copilot utility models (fallback/option)

Local Execution Behavior

When using a local Ollama-connected model, the emphasis is that:

  • Compute and logic run locally on the user’s machine (example mentioned: Mac mini)
  • The agent/model can access VS Code chat tools
  • Mentions possible integration such as an integrated browser

Agent Host Integration (Local Agent Harness + Copilot SDK Agent Host)

The speaker describes using:

  • A local agent harness
  • A Copilot SDK agent configured in a local agent host process

Two settings are called out as enabled:

  • chat.agentHost.byokModels.enabled
  • chat.agenthost.enabled

With agent host enabled:

  • The same model list becomes available inside the Agents window
  • This list includes:
    • Local Ollama models
    • Additional BYOK models (e.g., OpenRouter)

Main Speaker(s) / Sources

  • VS Code team (announced as the source of the “brand new updates”)
  • Unspecified presenter/speaker demonstrating the workflow inside VS Code

Original video