Live project proof
OnyxKraken AI
Self-Improving Autonomous Desktop Agent. A fully autonomous AI desktop entity with vision, voice, memory, knowledge retrieval, and self-improving code generation. It sees your screen, hears you, and acts independently.
Problem
Most agent demos stop at chat and tool calls. OnyxKraken investigates what happens when an AI system can see, remember, route tools, and act in a desktop environment.
Architecture
- Local-first agent stack with screen vision, voice input, memory, and FastAPI service boundaries.
- Capability routing connects perception, planning, tool execution, and feedback loops.
- Modular face pack, event bus, and Blender automation layers make behavior visible.
Evidence
- Live app: https://onyxkraken-face.netlify.app/
- Live hosted public face/app surface connected to the flagship system identity.
- Documented capabilities include YOLO vision, whisper.cpp voice, RAG memory, and 15+ API endpoints.
- Blender automation exercises show action beyond text output.
Tradeoffs
- Local autonomy requires careful permissions, inspectability, and safety boundaries.
- Desktop agents need environmental awareness before they can be trusted with larger workflows.
Outcome
The flagship proof system for MarkVizion: autonomous agent architecture, local AI, and creative tool control.
Stack
- Python
- FastAPI
- Ollama
- YOLO
- whisper.cpp
- React