Pinned Loading

1. llamacpp-gemma4-mtp llamacpp-gemma4-mtp Public

   Reproducible Gemma 4 multi-token-prediction bench harness for ik_llama.cpp. PR #1744 merged 2026-05-10. 2.6-2.98x lossless speedup verified. Two scripts to clone, build, bench.

   Shell 2

2. atelier-os atelier-os Public

   Multi-session AI desktop fleet. Sway+Wayland+wayvnc+noVNC. Per-employee container, iframe-embeddable, action-daemon for fast clicks. 259/259 tests. MIT.

   Python 1

3. destiny-computer destiny-computer Public

   A persistent Linux desktop that an AI owns. You watch it work. You take over the keyboard whenever you want. Your files persist between sessions. Docker compose. MIT.

   Python 1

4. moa-router moa-router Public

   Self-MoA (ICLR 2025) as a drop-in OpenAI-compatible proxy. Fan to 2 LLM upstreams, synthesize verbatim. +3.8%–6.6% on reasoning. 627 LOC FastAPI.

   Python

5. nandai-atelier nandai-atelier Public

   Destiny Atelier — single-file local AI chat. Three brains vote on every answer. 540 KB. 108 MCP tools. Open-source local Claude / GPT alternative. Local LLM. Runs on your own hardware.

   HTML

6. tooltalk tooltalk Public

   Drop-in middleware translating Gemma 4 text-format tool calls into OpenAI structured tool_calls. Streaming-safe, quote-tolerant. 972 LOC FastAPI.

   Python