Full-stack dev × AI. Running a local AI video production pipeline (2080Ti + MiniMax H3), with a habit of turning every pitfall I hit into documentation.
- Agent & RAG: shipped loci (queryable second brain · hybrid retrieval · MCP server), nacos2-mcp (MCP server for Nacos 2.x), sub-agent (multi-agent orchestration) and a function-calling tool loop in comfy-agent
- loci — A queryable second brain over scattered notes and docs — hybrid retrieval, cross-session memory, section-level citations, MCP server ·
pip install loci-rag - loci-dsh — Your loci second brain as a sidebar plugin in the DeepSeek Harness (dsh) web UI
- minimax-h3-turing — Field-tested handbook and compat workflows for running MiniMax H3 on a 2080Ti 22G (Turing sm_75)
- comfy-agent — Local-first visual studio for ComfyUI: pure-stdlib Python + vanilla JS, zero dependencies, ships as a single 10 MB exe
- nacos2-mcp — Read-only MCP server for Nacos 2.x — query configs and services from AI agents (fills the 2.x gap in official tooling)
- sub-agent — Full-flow AI coding orchestration: a 7-phase AGENTS.md that chains skills, tools and subagents
- desktop-pet — 3D chibi desktop pet, with a Studio that turns AI character sheets into new pets


