Gyaanmitra AI

Meet Gyaanmitra AI

The agent lives in the editor. It reads the files you have open, suggests a change, and waits for you before anything is written.

Gyaanmitra AI sits in the side bar, next to the code. It uses the model you connect with your own key. There is no hidden Gunamaya model in the middle. What leaves your machine is the API call you set up yourself.

Opening Gyaanmitra AI

1

Open the panel

Press Ctrl+Alt+I (Cmd+Ctrl+I on macOS), click the Gyaanmitra AI icon in the Activity Bar, or search "Gyaanmitra AI" from the Command Palette.

2

Connect a model

The first time you send a message, Gyaanmitra AI asks you to set up a model provider if you have not already. See Models & Providers.

3

Start a conversation

Ask a question, describe a change, or paste an error. The file you have open is included on its own. Type @ and a filename to pull in others.

What Gyaanmitra AI can do

  • Explain code you have not seen before, a stack trace, or how a change would ripple through the project.
  • Plan a change that takes several steps, and lay out the approach before it touches anything.
  • Edit files directly, and show every change as a diff you can review.
  • Run commands in the integrated terminal. It asks you before it runs anything that is not clearly safe.
  • Drive the Integrated Browser to click through a running app and confirm a fix actually works. See Web App Automation.

The settings surface

Gyaanmitra AI has its own settings, separate from the general editor settings. They are grouped into sections: Model (provider and API key), Tools & MCP (which tools and Model Context Protocol servers are available), Features (toggles for individual capabilities), Terminal (how shell commands run), Hooks (your own scripts on specific events), Plugins, Rules, Skills, and Subagents. Open them from the gear icon next to the chat input, or search "Gyaanmitra AI" in the Command Palette.

Not a chat extension

Gyaanmitra AI can use the editor, the terminal, and the Integrated Browser the way built-in features do. That is why it can run what it writes, and check it, instead of only describing a change.