Vigent

A Video Agent.

You describe what you want. It observes your project, calls the right tools — FFmpeg, vision models, transcribers, generators — and produces the output.

Pre-alpha. Built in the open.

What it can do today

• Understand video — analyze clips, extract content (Gemini 2.5 Pro)
• Transcribe — speech-to-text from audio / video (Gemini)
• Generate — images, video clips, voiceover (MiniMax)
• Search & fetch — web search and URL fetch as agent tools
• Shell — agent drives FFmpeg / curl / arbitrary pipelines
• Screen capture — macOS native screenshot and recording

What's next

• Frame-precise preview (WebCodecs in renderer)
• VS Code-style UI: file tree + video preview + agent chat
• Cut / overlay / retime / caption / render as agent tools
• Remotion project integration
• Project-aware agent memory

Stack

• Electron + React — desktop shell
• Node + pi-mono — agent loop
• Swift native — macOS screen capture
• FFmpeg — video processing
• Models — Claude (orchestration), Gemini 2.5 Pro (video / transcribe), MiniMax (image / video / TTS)

Requirements

• macOS 13+
• Node.js 20+, pnpm
• API keys: ANTHROPIC_API_KEY (required), GOOGLE_API_KEY and MINIMAX_API_KEY (for transcription and generation)

Getting started

pnpm install
pnpm -r build
pnpm dev:desktop

License

MIT