CODE · CONTEXT · AGENTS · REAL WORK
Making Real World Agentic — tool by tool, repo by repo.
Powered by Libraxis API
Tools, not features. Craft, not output. Proof, not promises.
Projects · The evidence
Tools for the craft of building with AI — and the work that proves them.
- 01
Code perception
It makes the repo actually visible. So the agent stops guessing and sees what breaks.
- 02
Agent context
Sessions, decisions, context, retrieval — so agent work has something to come back to.
- 03
Delivery layer
AI writes code. Vibecrafted closes the gap between a project and a product.
- 04
Real work
Where the thesis is tested: software work driven by dictation, veterinary ultrasound data, everyday writing. Built with the tools above, shipped in the open.
Tools in orbit
- rust-muxOne MCP server becomes a shared surface. Agents connect to the same tool, not ten copies of it.
- vc-operatorA local control room for Vibecrafted and your agents — terminal, MCP, tray, and Mac app in one place.
- mlx-batch-serverA local inference server for Apple Silicon. MLX models, batching, streaming, and OpenAI-compatible APIs on your Mac.
- vllm-swiftA vLLM Metal plugin powered by mlx-swift — LLM inference on Apple Silicon, no cloud required.
- mcp-server-semgrepSemgrep static analysis as an MCP server. Agents scan for vulnerabilities and code quality mid-conversation.
- unicode-puzzles-mcpGives agents a small Unicode puzzle box. They can hide, reveal, and verify messages inside ordinary text.