AI Agents Directory: 39 Agent Tools and Frameworks

An AI agent is a program that uses a language model to plan a task, call tools and keep going until the job is done, instead of answering one question and stopping. This page lists 39 agents and agent frameworks that are public, actively maintained and widely used. Every entry links to its own GitHub repository. Makers and licences were read from those repositories in November 2026, and the descriptions are ours. Star counts change fast, so we do not print them: follow the link on any card for the live numbers and latest updates. Our MCP servers directory covers the tools agents plug into.

Frameworks for building agents (13)

Libraries and toolkits for developers who write their own agents.

  • LangGraph

    Company project · langchain-ai

    A library for building agents as graphs of steps, with state you can pause, inspect and resume.

    Good for: Agents that need control over each step and a way to recover from failures.

    Licence: MIT

  • CrewAI

    Company project · crewAIInc

    A Python framework for teams of role-playing agents that split a job between them.

    Good for: Multi-agent workflows where each agent has a clear role.

    Licence: MIT

  • AutoGen

    Company project · microsoft

    Microsoft's programming framework for agentic AI, built around agents that talk to each other.

    Good for: Research and prototypes with several cooperating agents.

    Licence: CC-BY-4.0

  • OpenAI Agents SDK

    Company project · openai

    A lightweight OpenAI framework for multi-agent workflows with handoffs and guardrails.

    Good for: Python developers who want a small, readable core.

    Licence: MIT

  • Pydantic AI

    Company project · pydantic

    An agent framework from the makers of Pydantic, with typed outputs and tool use.

    Good for: Teams that care about validated, structured results.

    Licence: MIT

  • smolagents

    Company project · huggingface

    A tiny library where agents write and run code to solve tasks, from Hugging Face.

    Good for: Learning how agents work without a large framework.

    Licence: Apache-2.0

  • Google ADK

    Company project · google

    Google's code-first toolkit for building, evaluating and deploying agents.

    Good for: Developers on Google's stack who want evaluation built in.

    Licence: Apache-2.0

  • Mastra

    Company project · mastra-ai

    A TypeScript framework for AI applications with agents, workflows and memory.

    Good for: JavaScript and TypeScript teams.

    Licence: see repo licence

  • Agno

    Company project · agno-agi

    A Python platform for building and running agents with memory and tools.

    Good for: Shipping agents as a service without much glue code.

    Licence: Apache-2.0

  • CAMEL

    Community project · camel-ai

    A research-driven multi-agent framework that studies how agents cooperate at scale.

    Good for: Experiments with large groups of agents.

    Licence: Apache-2.0

  • Semantic Kernel

    Company project · microsoft

    A software kit for adding language model features and agents to apps in C#, Python and Java.

    Good for: Enterprise teams on the Microsoft stack.

    Licence: MIT

  • LangChain

    Company project · langchain-ai

    The widely used toolkit for connecting models to tools, data and agents.

    Good for: The broadest set of integrations in one place.

    Licence: MIT

  • LlamaIndex

    Company project · run-llama

    A platform for turning documents into data agents can search and reason over.

    Good for: Agents that must answer from your own files.

    Licence: MIT

Coding agents (10)

Agents that read your project, edit files and run commands.

  • Claude Code

    Company project · anthropics

    Anthropic's coding agent for the terminal. It reads your project, edits files and runs commands.

    Good for: Developers who like working from the command line.

    Licence: see repo licence

  • Codex CLI

    Company project · openai

    OpenAI's lightweight coding agent that runs in your terminal.

    Good for: Quick code tasks and refactors from the command line.

    Licence: Apache-2.0

  • Gemini CLI

    Company project · google-gemini

    Google's open source agent that brings Gemini into the terminal.

    Good for: Coding and file tasks with Gemini models.

    Licence: Apache-2.0

  • Cline

    Company project · cline

    An autonomous coding agent for your editor or command line that asks before it acts.

    Good for: VS Code users who want step-by-step approval.

    Licence: Apache-2.0

  • Continue

    Company project · continuedev

    An open source coding agent and assistant that works inside your editor.

    Good for: Teams that want to choose their own models.

    Licence: Apache-2.0

  • OpenHands

    Company project · OpenHands

    An open source platform of agents that write code, run it and fix it in a sandbox.

    Good for: Delegating whole development tasks.

    Licence: MIT

  • opencode

    Company project · sst

    An open source coding agent for the terminal that works with many model providers.

    Good for: People who want a terminal agent without being tied to one model.

    Licence: MIT

  • SWE-agent

    Research project · SWE-agent

    Takes a GitHub issue and tries to fix it on its own. Born in academic research on software agents.

    Good for: Studying how well agents handle real bugs.

    Licence: MIT

  • goose

    Company project · block

    An extensible open source agent that goes beyond code suggestions: it installs, runs, edits and tests.

    Good for: Automating developer chores on your own machine.

    Licence: Apache-2.0

  • Open Interpreter

    Company project · OpenInterpreter

    A coding agent that runs code locally, aimed at open models.

    Good for: Local, model-agnostic code execution.

    Licence: Apache-2.0

Browser and computer agents (5)

Agents that use a web browser or a desktop the way a person would.

  • Browser Use

    Company project · browser-use

    An open source Python library that lets agents use a web browser, with a hosted cloud option.

    Good for: Automating web tasks that have no API.

    Licence: MIT

  • Skyvern

    Company project · Skyvern-AI

    Automates browser workflows with AI, so forms and portals can be handled without brittle scripts.

    Good for: Repetitive tasks on sites that change often.

    Licence: AGPL-3.0

  • Stagehand

    Company project · browserbase

    An SDK for extracting data from and acting on any website, mixing code with plain language.

    Good for: Developers who want reliable browser automation.

    Licence: MIT

  • Nanobrowser

    Community project · nanobrowser

    An open source browser agent that works inside your own logged-in Chrome.

    Good for: Web tasks on accounts you are already signed into.

    Licence: Apache-2.0

  • UI-TARS Desktop

    Company project · bytedance

    ByteDance's open source multimodal agent stack for controlling computers and browsers from screenshots.

    Good for: Experimenting with computer-use agents.

    Licence: Apache-2.0

Workflow and no-code builders (4)

Visual platforms for wiring models, tools and data into working apps.

  • n8n

    Company project · n8n-io

    A workflow automation platform with AI features. Its fair-code licence is source-available, not fully open source.

    Good for: Connecting apps and agents with a visual editor.

    Licence: see repo licence

  • Langflow

    Company project · langflow-ai

    A visual builder for agents and workflows, which you can then deploy.

    Good for: Building agent flows without much code.

    Licence: MIT

  • Dify

    Company project · langgenius

    A platform for agentic workflows and retrieval pipelines with a visual builder and many model options.

    Good for: Teams that want one place to build and run AI apps.

    Licence: see repo licence

  • AnythingLLM

    Company project · Mintplex-Labs

    A private, all-in-one app for chatting with your documents and running agents, on desktop or self-hosted.

    Good for: Keeping data on your own machine.

    Licence: MIT

Research and knowledge agents (3)

Agents that search, read and write up what they find.

  • GPT Researcher

    Community project · assafelovic

    An autonomous agent that researches a topic across the web and writes a sourced report.

    Good for: Quick first-draft research with citations.

    Licence: Apache-2.0

  • Khoj

    Company project · khoj-ai

    A self-hostable AI second brain that answers from the web and your own notes.

    Good for: Personal knowledge search.

    Licence: AGPL-3.0

  • DeerFlow

    Company project · bytedance

    ByteDance's open source agent harness for long tasks that research, code and create.

    Good for: Long, multi-step research jobs.

    Licence: MIT

Memory and open standards (3)

The pieces that help agents remember and work together.

  • Mem0

    Company project · mem0ai

    A memory layer that lets agents remember users and past conversations.

    Good for: Making an assistant feel consistent over time.

    Licence: Apache-2.0

  • Letta

    Company project · letta-ai

    A platform for stateful agents with long-term memory that can learn and improve.

    Good for: Agents that must carry context across sessions.

    Licence: Apache-2.0

  • Agent2Agent (A2A) protocol

    Open standard · a2aproject

    An open protocol for agents from different makers to talk to each other.

    Good for: Connecting agents built with different tools.

    Licence: Apache-2.0

General-purpose agents (1)

Agents built to take on open-ended tasks.

  • AutoGPT

    Company project · Significant-Gravitas

    One of the first autonomous agent projects, now a platform for building and running agent workflows.

    Good for: Exploring what general-purpose agents can do.

    Licence: see repo licence

How to choose an agent

  • Decide the job first: writing code, browsing, research or a business workflow. Each group above fits a different job.
  • Start with the smallest tool that does it. A framework gives control but asks for code, a ready-made agent asks for trust.
  • Check what it is allowed to touch. Agents can edit files, spend money and send messages, so limit access and keep approvals on.
  • Read the repository: recent updates, answered issues and a licence you can live with. Some projects use source-available licences instead of open source ones.
  • Test on something low risk before you point it at real work.

Questions people ask

What is the difference between an agent and a chatbot?

A chatbot answers. An agent works toward a goal across several steps, deciding which tools to use along the way, such as a browser, a code editor or a database.

Are these agents free?

Most of the code here is free to download and run. You still pay for the language model behind it, and some projects also sell a hosted version.

Which one should a beginner start with?

If you code, a terminal or editor agent from the Coding group is the quickest way to see the idea working. If you do not code, try a visual builder from the Workflow group.

Are agents safe to run on my own computer?

They can be, with care. Run them in a sandbox or a test project first, give them narrow permissions and read every action they ask to take.

How do agents connect to other tools?

Often through the Model Context Protocol. See our MCP servers directory for servers an agent can use.

Know an agent that belongs here? Tell us and we will check it before listing it.