Two students burned Karpathy's microGPT into FPGA fabric, generating 50,000+ tokens per second with zero Python, zero CUDA, and zero DRAM in the inference path.
A one-click portable ComfyUI installer that handles Nunchaku, SageAttention, FlashAttention and CUDA/PyTorch version swaps without touching a terminal.
An open-source desktop app called Agent Orchestrator lets you run 26 different coding agents in parallel across Git worktrees, with a live Kanban tracking every PR.
Gondolin runs agent-generated code inside disposable Linux micro-VMs while the host mediates every network call and injects secrets the guest never sees.
Vibe is a free, open-source desktop app that runs OpenAI Whisper locally on your GPU — no cloud, no cost, no data leaving your machine — across Windows, macOS, and Linux.
A minimalist open-source project runs Claude agents inside Docker containers, connects them to messaging apps like Slack and WhatsApp, and swaps configuration for code edits.
Mozilla.ai's cq is an open protocol that lets coding agents share hard-won lessons across sessions, users, and tools instead of rediscovering the same bugs.
EBM-RL teaches vision-language models to see a scene, think about it, then speak in character — unlocking immersive NPC dialogue for VR and interactive storytelling
SPARK introduces PDI, a trajectory-level metric that tells you whether an AI agent's distilled skills are grounded in real environment evidence — and uses it to make smaller, cheaper models outperform their expensive teachers
REFLECTOR trains LLMs to catch themselves mid-generation using a two-stage SFT and RL pipeline, blocking indirect jailbreaks with over 90% success while actually improving model utility.