JeroenPieksma/claude-computer
Claude Computer demonstrates AI autonomy in a virtual machine with real-time streaming, research, creation, and exploration. Watch Claude navigate, interact, and learn in real time 🐙
What's novel
Claude Computer demonstrates AI autonomy in a virtual machine with real-time streaming, research, creation, and exploration. Watch Claude navigate, interact, and learn in real time 🐙
Code Analysis
10 files read · 3 roundsA FastAPI + Next.js platform that wraps Anthropic's computer-use API to stream Claude's autonomous desktop interactions (driven by hardcoded Gen Z crypto-slang prompts) to a live web viewer with WebSocket-based real-time screen and activity streaming.
Strengths
The computer-use-demo sub-project is a well-structured, functional implementation of Anthropic's computer-use tools with proper coordinate scaling, tool versioning, and prompt caching. The backend service layer (screen capture, activity logging, behavioral system) is cleanly separated with reasonable async patterns.
Weaknesses
The 'autonomous' behaviors are trivially thin — they just pick random strings from hardcoded lists and send them as chat messages to Claude, with no real decision-making, state tracking, or feedback loops. The README heavily oversells the project with 'CCFTokens', crypto incentives, and 'advanced ML pipeline' language that doesn't exist in the code. The Gen Z crypto persona is a gimmick, not a feature.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
12
Contributors
2
Files
97
Active weeks
2
Repository
Language
Python
Stars
2
Forks
0
License
MIT