tonbi

turboquant-pytorch. From-scratch PyTorch implementation of Google's TurboQuant (ICLR 2026) for LLM KV cache compression. 5x compression at 3-bit with 99.5% attention fidelity.

hermes-multi-agent-workflow. Python

llm-wiki. Open-source template for building LLM-powered knowledge bases following Karpathy's LLM Wiki pattern

model-trader. Python

hermes-canvas. Frontend UI Studio built into the Hermes Agent Web Dashboard

hyperframes-prompts.

learn-ml. 12-episode video series teaching ML engineering by building real projects

moe-ssd-streaming-windows. Running a 32 GB AI model on 28 GB of memory — MoE expert streaming from NVMe SSD on Windows

open-video-studio. A local-first studio for generating and comparing AI videos through OpenRouter video models.

hisa-pytorch. Pure PyTorch implementation of HISA (Hierarchical Indexed Sparse Attention) as a drop-in plugin for HuggingFace models

hermes-office. JavaScript

aigakuen. Create hyper-specialized Otaku agents with deep domain expertise for Claude Code

gqa-flash-attn-rtx3060. Triton GQA flash attention kernel for NVIDIA RTX 3060 (sm_86). Matches or beats flash-attn2 on 9/10 workloads; up to 2.95x faster on long-context decode.

hermes-dashboard-wallet. Python

agentic-abstention-mini. Python

leakplug-public. TypeScript

automaton. The first AI that can earn its own existence, replicate, and evolve — without needing a human

agentwikis-mcp. MCP server for Agent Wikis — maintained, agent-readable knowledge bases. Read-only, no API key.

ClawMPP. Machine Payments Protocol integration for OpenClaw — one wallet, every AI service, no API keys

llamacpp-autotune. Python

agentpay-solana. AgentPay SDK for Solana — self-custodial, policy-aware wallet operations for AI agents

nuke-shelter. Solana memecoin supply distribution analyzer — traces holder acquisitions back to original DEX swaps

clarityx402. TypeScript

mini-vss. Desk-scale study of the NVIDIA VSS loop: retrieve-then-verify video search on one GPU and a Claude Code session