The whole shelf
Every open-source building block in the atlas, in one flat list. Search by name or tag, or filter by category.
A2A
An open protocol that lets independent agents discover each other and collaborate across frameworks and vendors.
view source ↗AgentShield
Adds guardrails and security checks around agents to catch unsafe or unintended actions.
view source ↗AirLLM
Run 70B-parameter LLMs on a single 4GB GPU — AirLLM does layer-by-layer inference so huge models fit on modest hardware, with no quantization, distillation, or pruning required.
view source ↗browser-use
Gives an agent control of a real browser so it can click, type, and navigate the web like a person.
view source ↗ccusage
A small CLI that breaks down your Claude Code token usage and spend.
view source ↗Claude Skills collection
A collection of reusable Claude skills you can plug in for common tasks.
view source ↗Composio
A tool layer that wires agents into hundreds of real apps and APIs with managed authentication.
view source ↗Context7
Feeds version-correct documentation for the libraries you're using straight into the model's context, via MCP.
view source ↗DeepAgents
Scaffolding for long-horizon agents that plan, spin up sub-agents, and keep working across many steps.
view source ↗DeepSeek-OCR
A vision model that reads documents and compresses pages into very few tokens, making long PDFs cheap to feed a model.
view source ↗DeepWiki-open
Point it at any GitHub repo and it generates a browsable, navigable wiki that explains the codebase.
view source ↗Dolphin
Parses document images — tables, columns, figures — into clean structured layout for downstream use.
view source ↗Firecrawl
Crawls entire sites and returns clean, LLM-ready markdown or structured JSON — the scraping workhorse behind many RAG pipelines.
view source ↗Jina Reader
Point it at any URL and it hands back clean, model-ready text, stripping the page down to what matters.
view source ↗LangExtract
Uses an LLM to pull structured fields out of messy text, keeping each value linked back to where it came from.
view source ↗Langfuse
Open-source observability for LLM apps: trace every call, run evals, and watch cost and quality over time.
view source ↗MCP Registry
The official directory of Model Context Protocol servers — a package index for agent tools.
view source ↗NLWeb
Turns a website's own content into a conversational, agent-readable interface.
view source ↗OpenHands
An open platform for software-building agents — hand it a task and it plans, writes code, runs commands, and browses the web until the work actually ships.
view source ↗OpenSpec
A spec-first workflow that keeps AI coding aligned to an agreed, reviewable specification.
view source ↗power-pwn
A red-team toolkit for probing Microsoft 365 Copilot and Power Platform — for security research, not production.
view source ↗Presidio
Detects and redacts personal data — names, emails, IDs — in text and images; a natural partner to PromptMask.
view source ↗PromptMask
Local-first privacy layer for LLMs that redacts sensitive data before cloud API calls and restores it in responses.
view source ↗Pydantic-AI
An agent framework from the Pydantic team that brings type-safe, validated outputs to tool-calling agents.
view source ↗RAGFlow
Open-source RAG engine with deep document understanding, agent workflows, and production-ready ingestion pipelines.
view source ↗ScrapeGraphAI
Python library for AI-powered web scraping — describe what you want extracted and let an LLM-driven graph pipeline handle the rest.
view source ↗Spec Kit
GitHub's toolkit for spec-driven development — write the spec first and let agents build to it.
view source ↗Subagents collection
A ready-made library of specialized sub-agents you can drop into your own agent setup.
view source ↗Temporal
A durable execution engine that keeps long-running agent workflows reliable through retries, failures, and restarts.
view source ↗UI-TARS Desktop
A GUI agent app that operates your actual computer — apps and windows — from natural-language instructions.
view source ↗