Explore the directory

Awesome Scientific Language Models is a curated list of pre-trained language models for scientific domains such as mathematics, physics, chemistry, materials science, biology, medicine and geoscience, accompanying a survey paper on scientific large language models.

• Papers

Copilot Developer Camp is a virtual workshop of hands-on labs for developers who want to build agents and connectors for Microsoft 365 Copilot, from no-code declarative agents to custom engine agents with full control of the model and orchestrator.

• Tutorials

A low-latency, streaming automatic speech recognition model that commits transcribed text permanently as it arrives, without revising earlier words, for uses such as live captioning and voice-driven agents.

• Speech

SkillCorpus is an open-source pipeline that turns scattered SKILL.md files from public repositories into a curated, retrieval-ready corpus of agent skills, with safety and license gating, quality evaluation, and task-specific skill matching built in.

• Agent skills
• RAG

obsidian-mcp-server is an MCP server that reads, writes, searches and surgically edits Obsidian vault notes, sections, frontmatter and tags over Obsidian's Local REST API plugin, with folder-scoped read and write permissions.

• MCP servers

Minke is a local-first desktop agent workspace, powered by DeepSeek Harness, that brings conversations, files, terminals and a shared browser into one app for macOS, Windows and Linux.

• Coding agents

A standalone C++ inference engine for real-time, multi-person 3D body pose reconstruction from a single camera, exporting motion-capture BVH files or MPEG avatar containers with no Python dependency at runtime.

• Inference

Cheetah is an on-device streaming speech-to-text engine from Picovoice that transcribes audio locally in real time, without sending voice data to the cloud.

• Speech

AINL is a compact, graph-canonical programming system for building AI workflows that need multiple steps, state and memory, tool use, repeatable execution, and validation, instead of relying on long prompt loops.

• Agent frameworks

RAGLight is a lightweight, modular Python framework for Retrieval-Augmented Generation that lets a developer plug in different LLMs, embedding models and vector stores, with built-in MCP integration for external tools and data sources.

• RAG

SandBase Harness is a local-first, self-hosted AI agent runtime and MCP bridge that runs sandboxed agent sessions, memory, credential vaults and audit trails on the user's own machine or infrastructure, with a built-in web console.

• Agent frameworks
• MCP servers

AgentOS is an open-source TypeScript framework for AI agents with persistent cognitive memory, runtime tool forging, multi-agent orchestration and one dispatch interface across 11 LLM providers.

• Agent frameworks

AI Memory Vault is a free, open-source system and template set that turns an Obsidian vault into an AI coding agent's persistent working memory, using plain markdown files instead of a vector database.

• Agent memory

MetaHarness mints a custom, branded AI agent harness for any GitHub repository: its own CLI, MCP server, project-scoped memory, recommended agents and skills, and a governance policy, generated in under a minute.

• Agent frameworks

Switch is open-source, self-hostable infrastructure that connects AI agents from any provider or framework to the messaging tools a team already uses, such as Slack, Microsoft Teams, Discord, Telegram and Mattermost.

• Workflow automation

wcgw is an MCP server that gives Claude and other MCP clients an interactive shell and code-editing tools on the local machine, with unrestricted, unfiltered access to the user's files and commands.

• MCP servers

MaverickMCP is a personal-use FastMCP server that gives any MCP client, such as Claude Desktop, Claude Code, Cursor or VS Code, tools for stock data analysis, technical indicators, stock screening and portfolio tracking, running entirely on the user's own machine.

• MCP servers

DSH Desktop is an out-of-the-box Windows and macOS desktop client that bundles the DeepSeek Harness (dsh) CLI and all of its official plugins with an embedded Node.js runtime, so it launches with a double-click and needs no separate Node.js install.

• Coding agents

RAG Me Up is a simple, extensible framework for building Retrieval-Augmented Generation applications quickly, with a Docker Compose stack covering a Postgres database, a Node.js API server, a React UI and a Python RAG server.

• RAG

cavemem gives coding assistants cross-agent persistent memory: hooks capture session observations, compress them into a local SQLite store, and expose them back through MCP tools. It is now frozen; its memory core continues in a successor, caveman.

• Agent memory

myAiVideos is an open-source, automated short-video production pipeline that takes a topic in and produces a publish-ready Chinese-language video, coordinating multiple AI models across topic discovery, script generation, visuals, audio and post-production.

• Generative media

A vector post-training quantization method that compresses large language models to under two bits per parameter while keeping them usable, with pre-quantized community models available for models as large as Llama 3.1 405B.

• Models
• Inference

An open-source tool that converts a web design screenshot, sketch, wireframe, or Figma or XD export into clean, responsive HTML, CSS, and JavaScript, using GPT-4 vision.

• Design tools

SwarmClaw is a self-hosted, open-source AI agent runtime and multi-agent framework for running autonomous agents and agent swarms, with durable memory, MCP tools, scheduling and delegation across more than 23 LLM providers.

• Agent frameworks