Explore the directory

DocsAgent is an MCP server that gives AI agents fast, private, local full-text search over a personal Zotero library, using a native C++ BM25 search engine so results return in milliseconds without sending documents anywhere.

• MCP servers

A fork of vLLM, the high-throughput LLM inference and serving library, adding an experimental TurboQuant key-value cache path tuned for specific CUDA workstation GPUs.

• Inference

Grace is a domain-specific functional programming language for prompting language models, with a built-in prompt function and a type checker that can infer the JSON schema an output should follow from how the result is used.

• Languages

AOCI-CODE maintains a persistent, Git-versioned index of a codebase's files and database schema, written by the coding agent itself, so agents such as Claude Code or Cursor can read the index once instead of re-exploring the repository on every task.

• Agent memory

Prowl is a native macOS terminal built for running several AI coding agents such as Claude Code and Codex in parallel, with a live overview of every running agent, a stacked view of worktrees, and a command-line interface agents can use to drive the terminal themselves.

• Coding agents
• Terminals

Open-AgentRL is an open-source collection of reinforcement-learning research for large language model agents, including a system that jointly trains policy and reward models, a study of what makes agentic RL training effective, and a method for dynamic tool selection.

• Models

KaliGPT is a command-line agentic assistant for Kali Linux that uses Gemini, ChatGPT, Ollama or OpenRouter models to help with ethical hacking and cybersecurity learning and workflows.

• Security

pstack ports Lauren Tan's opinionated Cursor skill stack, poteto-mode, to Claude Code, Codex and other agent harnesses, routing a stated goal to the right workflow to keep changes concise, simple and independently verified.

• Agent skills

Pi Extensions is a monorepo of independently installable extensions for the Pi coding agent, covering coding and delegation, research, browser automation, workflow management, observability and terminal ergonomics.

• Agent skills

Vendo is an embedded AI agent that B2B products add so their own customers can ask for new views or automations, which the agent builds live from the product's existing API and renders inside a sandboxed, brand-matching surface without touching the product's source code.

• Agent frameworks

Neon MCP Server is an open-source MCP server that lets an AI agent manage Neon Postgres databases in natural language, creating projects and branches, running queries and performing migrations.

• MCP servers
• Databases

Nubase is a self-hostable, AI-native backend platform that an AI coding agent can drive directly over MCP to model data, deploy backend logic, publish a generated frontend and schedule recurring work, combining database, auth, storage and memory in one service.

• MCP servers
• Backend

Corbell builds a living architecture graph of a multi-repository backend system from its code and past design documents, then uses that graph to generate specs that follow established patterns and to review proposed changes.

• Agent memory

A curated, actively updated list of papers, surveys, frameworks and posts about machine unlearning in large language models, that is, removing specific knowledge or training data from a model after the fact.

• Papers

An automatically updated overlay of pre-built, often unreleased, packages for Wayland compositor tools such as sway and wlroots, for the NixOS unstable channel, with a binary cache so packages do not need to be rebuilt locally.

• CI/CD

SINQ is a calibration-free quantization method that shrinks large language models to lower precision while keeping most of their accuracy, quantizing a 14B model in about 21 seconds and a 236B model in about 5 minutes.

• Inference

Shelley is a mobile-friendly, web-based coding agent for exe.dev that supports multiple simultaneous conversations, multiple models and multimodal input such as screenshots, and leaves authorization and sandboxing to whoever runs it.

• Coding agents

Codex MCP Server is a bridge that lets Claude Code call OpenAI's Codex CLI as an MCP tool, for AI-assisted code analysis, generation, review and web search from inside another agent's session.

• MCP servers

A tool that drives the Codex CLI in a continuous review-and-reconcile loop toward a stated goal, running each round in a clean context so earlier attempts do not pollute later ones.

• Coding agents

SonarQube MCP Server is an official Model Context Protocol server that connects an AI coding agent to SonarQube Server or SonarQube Cloud for code quality and security analysis, including snippets analyzed directly in the agent's context.

• MCP servers
• Code quality

llms.py is a lightweight command-line tool, API and web chat interface for using multiple large language models entirely offline, keeping all data in the browser's own storage rather than any server.

• Inference

Relax is an asynchronous reinforcement-learning framework for post-training large multimodal language models at scale, decoupling training and inference so agents can learn from text, image, video and audio interactions.

• Models

pi-acp is an adapter that lets the pi coding agent speak the Agent Client Protocol over stdio, so editors such as Zed can drive pi as an ACP agent with streamed output, tool-call locations and structured diffs.

• Coding agents

sqz is a Rust command-line tool that compresses tool output AI coding agents send into the model's context, such as build logs and repeated file reads, while leaving source code, stack traces and secrets untouched, and can restore any result exactly.

• Agent memory