Toolbox
A live catalog of AI agent harnesses, rebuilt daily from GitHub data. We measure attention, not quality.
Rankings: Most starred · Fast growing · Newest · Most wanted
Browse by category
We group harnesses by the job they do. Pick a category to see every project in it, most starred first.
Coding agents
Frameworks & SDKs
general-purpose libraries for building agents
coordinating teams, crews and swarms of agents
graph, DAG and state-machine runtimes for agent workflows
visual and no-code agent builders
sandboxes and code execution environments for agents
Agent capabilities
memory layers and context management
retrieval and knowledge layers for agents
MCP servers, tool registries and function-calling plumbing
agents that operate a desktop or GUI
browser automation and web agents
voice and realtime conversational agents
Running agents in production
Applied agents
deep research and web research
data analysis, BI and SQL agents
agents that run training experiments, tune and evolve models
agents that plan, code and debug for robots and simulators
assistants and chat-based task agents
Methodology
We find candidates through GitHub Search (agent-related topics and keywords, plus a radar for repos younger than 90 days), then snapshot every repo in the pool once a day via the GraphQL API. For new entries we rebuild star history from sampled stargazer timestamps.
Categories come from the job a project does, not the language it is written in. Every project sits in exactly one, and within a category we sort by stars.
Exclusions:archived repositories, inactive repos (no pushes in 60 days), awesome-lists, and prompt collections. We define a “harness” as a tool or framework that orchestrates an LLM to complete tasks — tool calling, context management, agentic loops. Models themselves, API wrappers, and prompt libraries do not qualify.