Explore the directory

An Apache-2.0 toolkit for training a custom wake word and running it fully on-device in Python, the browser, or on iOS and Android.

• Speech

A ComfyUI node that expands a short prompt into a full styled prompt, negative prompt and aspect ratio for Qwen-Image-2.1, using one of 17 photographic styles.

• Generative media

A design CLI driven from Claude Code, Codex or Antigravity: describe a UI in plain language and get web, Figma or SwiftUI output checked by deterministic quality gates.

• CLI tools
• Design tools

A Swift package for streaming, on-device speech recognition on iOS, built on a custom Rust inference runtime (VoxRT) rather than ONNX Runtime or PyTorch Mobile, running a 32-million-parameter FastConformer model with roughly 90ms per-chunk latency on an iPhone 13 Pro Max.

• Speech
• Mobile

A fork of vLLM tuned to run mixture-of-experts models several times larger than the available memory on a single NVIDIA GB10 machine (DGX Spark, ASUS Ascent GX10), using 2-bit expert planes with hand-written SASS kernels, an FP4 quality-recovery tier and speculative decoding.

• Inference

A coding agent that lets a developer write Python functions as tools, which the agent then chains, runs in parallel and inspects before deciding what to fold back into the conversation, following the same session across terminal, desktop and phone clients.

• Coding agents

An open-source monorepo of 31 production-focused MCP servers for services such as GitHub, Slack, Notion, databases and cloud platforms, sharing one strict-mode TypeScript core, one CLI for install and diagnostics, and consistent validation, logging and transport support.

• MCP servers

A local-first command-line tool for semantic code search that indexes functions, methods, types and their attached documentation with Tree-sitter, then finds them by meaning rather than exact name, using local GGUF embeddings by default and sqlite-vec for storage.

• RAG
• CLI tools

A near-lossless 5-bit transformer compression method with verifiable, reproducible reconstruction: a CLI generates text against a hosted compressed model, browses a model catalog, and verifies a downloaded pack's integrity against a published SHA-256 manifest.

• Models
• Inference

An AI coding assistant that runs as both a command-line tool and a Visual Studio Code extension, using a bring-your-own-key model where the developer pays their configured AI provider - AWS Bedrock or OpenRouter - directly instead of a subscription.

• Coding agents
• Editors

An MCP server that lets a coding agent generate synthetic training datasets, run LoRA fine-tuning, do policy-gradient reinforcement learning against a local judge model, merge adapters and evaluate results, all without leaving the chat.

• MCP servers
• Models

A production-grade vector and graph database engine written in Rust, where a vector's identity and its graph node are the same object, so a single query can combine nearest-neighbor similarity search with explicit, typed graph traversal without syncing two separate databases.

• Databases

A cross-platform desktop app that adds a voice interface to coding agents such as Claude Code, Cursor, Codex and Gemini CLI: you talk to the agent out loud and hear its answers in a natural voice, with everything running on the local machine.

• Coding agents
• Speech

An open protocol and TypeScript reference stack that authorizes and logs AI agent actions with real-world consequences, using Ed25519 capability tokens, tiered human-approval rules and a tamper-evident, hash-chained evidence log.

• Security
• Auth

A semi-structured specification format, now at version 0.6, for describing work assignments for autonomous agents: a document states required behavior, completion criteria, examples and failure handling, checked by a deterministic Go linter.

• Agent frameworks

Research code for a Blackwell-GPU LLM inference method that combines a sparse FP4 main computation path with a dense FP4 residual path to recover accuracy lost to sparsification and quantization, with a PyTorch-only simulation mode for accuracy testing without the CUDA kernel.

• Inference
• Papers

A zero-dependency, single-file inference engine for Qwen 3.5 language models written in pure Java, supporting model sizes from 0.8B to 35B (including a mixture-of-experts variant), several GGUF quantizations, and GraalVM native-image compilation.

• Inference

A starter kit of step-by-step instructions, not code, for building a shared program-memory layer for AI agents using Claude Code, native MCP integrations and git, covering the pull, map and loop levels of the architecture.

• Agent memory
• Tutorials

An AI agent framework for the openvela embedded operating system, running on watches, glasses, speakers and similar devices, with multiple LLM backends, a ReAct tool-calling loop, an MCP client, and configuration tuned for devices with as little as 256KB of RAM.

• Agent frameworks

An OpenCode plugin that gives coding agents persistent memory using a local SQLite and usearch vector database, auto-capturing facts from conversations and exposing search, profile and list operations as an agent tool, with no cloud service required.

• Agent skills
• Agent memory

A lightweight runtime bus that bridges MCP servers, Matrix rooms and HTTP/SSE endpoints into one set of Invoke, StreamInvoke and Publish/Subscribe flows, giving heterogeneous agents cross-process, cross-language collaboration with session and trace propagation.

• Agent frameworks

A standalone Rust MCP server that gives AI coding agents filesystem tools - viewing, editing, batch-editing and searching files, running shell commands, and transparent SSH/SFTP access - through a single binary with no separate runtime.

• MCP servers
• CLI tools

An MCP server and CLI for semantic search over an Obsidian vault or any folder of linked Markdown files, generating embeddings through OpenAI, Ollama or LM Studio and storing vectors locally in sqlite-vec with no telemetry.

• MCP servers
• RAG

A Go-based MCP server that dynamically exposes the tasks defined in a project's Taskfile.yml as individual MCP tools, letting an AI assistant discover and run build, lint and other project tasks the same way a human developer would.

• MCP servers