Chat with Claude, Gemini, Grok, ChatGPT, DeepSeek, Qwen, and Llama in one shared VS Code session. Every configured agent silently receives the full conversation as context, but only the agent(s) you @-mention actually respond.
@Claude fix the failing test in utils.ts — Claude alone runs a full agentic tool-use loop (read/write/edit/grep/glob/bash/web) against your workspace, with permission prompts and plan-mode for non-trivial work. Every other agent stays silent but stays caught up.
@Claude @Gemini what's the best approach here? — Claude and Gemini discuss back and forth in plain text, visible in the transcript, with no file access at all.
During a multi-agent discussion, any agent can propose a concrete plan dividing up the work; once you approve it, the assigned agents run their own agentic loops concurrently.
Setup
npm install
npm run compile
Press F5 to launch the Extension Development Host.
Run Multi-Agent Chat: Set API Key… to configure whichever providers you want to use — each one is optional and independent.
Click the Multi-Agent Chat icon in the top-right of the editor title bar (or run Multi-Agent Chat: Open Chat) to open the panel.
Each agent talks directly to its own provider's API using a key you supply, stored in VS Code's SecretStorage — never in .env or settings.json. See .claude/plans (or your own plan doc) for full architecture notes.
Development
npm run watch — rebuild on change (used automatically by the F5 launch config).
npm run lint — ESLint over src/ and webview-ui/.
npm test — unit tests that don't require a real VS Code instance.
