LibreAI ENESPTDEFRRUJA
← All categories// agents

Best Open-Source AI Agent Tools

Autonomous AI agents and multi-agent frameworks.

479 tools · 💎 105 hidden gems

// start here

Start here

The most-starred projects in this category, plus a rotating pick of three more from deeper in the list — refreshed so the whole catalogue gets its turn in the window.

AI Agents

openclaw

A personal AI that really does things — on any OS, with your data staying yours.

★ 391.6k starsMIT
View guide →
AI Agents

superpowers

An agentic skills framework & software development methodology that works.

★ 294.9k starsMIT
View guide →
AI Agents

ECC

Agent harness optimization system with skills, memory and security for Claude Code, Codex and more

★ 275.2k starsMIT
View guide →
AI Agents

adk-js

Google's code-first TypeScript toolkit for building AI agents.

★ 1.4k starsApache-2.0
View guide →
AI Agents

duckle

Open source ETL/ELT you deploy on your own servers or cloud.

★ 1.4k starsApache-2.0
View guide →
AI Agents

tradememory-protocol

Tradememory-protocol — decision audit trail + persistent memory for ai trading, open-so...

★ 1.4k starsMIT
View guide →

More categories: VideoImageAudioAI MusicAI 3DClippingLocal LLMAI ChatRAG & KnowledgeAutomationProductivityDev ToolsDesignWritingSelf-HostedEasy Install

// about this category

About Best Open-Source AI Agent Tools

AI agents go beyond chat: they plan multi-step work, call tools, browse, write and run code, and report back. This category spans single-agent frameworks, multi-agent teams where specialised agents collaborate, computer-use agents that operate a desktop, and the harnesses that keep them observable and safe.

The honest dividing line is how much supervision each system needs. Some are research frameworks for building your own agent; others are ready-made workers for coding, research, or browser tasks. Every guide here is explicit about what the agent can actually do today, because this is the category where demos most often outrun reality.

How to choose

  • Start with a narrow, checkable job — codebase questions, research summaries — before giving an agent open-ended work.
  • Prefer frameworks with visible traces and approvals over black boxes; you will debug the agent's reasoning, not just its output.
  • Sandboxing matters: agents that run code should do it in a container or VM, and the good ones say so.
// faq

Questions, answered

Which open-source AI Agents tools are listed here?

This page collects 479 open-source AI Agents projects curated by LibreAI — Autonomous AI agents and multi-agent frameworks. Each one is verified on GitHub and has its own guide page with an overview, install steps, use cases, and an FAQ, so you can judge fit, requirements, and licensing before you download anything.

Are these AI Agents tools free to use?

Yes. Every tool here is open source and free to download and use under its own license — most use MIT, Apache-2.0, or GPL-3.0, and any custom source-available license is flagged on the tool's page. Check the LICENSE file before commercial use.

Does LibreAI host these tools?

No. LibreAI publishes guides only — no downloads are served from this site, and no tool here is mirrored or repackaged by us. Every install comes from the project's official GitHub repository, which each guide links to directly, so you always get the genuine, current release from the people who build and maintain the tool.

How do I install a AI Agents tool from this page?

Open the tool's guide and follow its 'Run it yourself' section — the steps come from the project's own documentation — or use the 'View on GitHub' button to go straight to the official repository. Each guide also states the tool's requirements first, so you can check your machine qualifies before downloading anything.

How were these tools chosen?

Every project is verified live through the GitHub API — stars, forks, license, and recent activity — and listed only if it has real momentum and an active community. Projects are re-checked as the catalogue is refreshed, and any project that goes quiet or changes its licensing terms has that noted on its page.

Can AI agents really work unsupervised?

Short, well-defined jobs — answering questions about a codebase, drafting research summaries — work well today. Long open-ended tasks still drift, which is why the better frameworks here emphasise traces, checkpoints, and human approvals rather than full autonomy. Plan on reviewing early runs closely; trust is earned per task type, not granted to the framework as a whole.

Are agents safe to give my computer or accounts to?

Treat an agent like a new contractor: start with limited access, sandboxed environments, and review of consequential actions. The projects in this category differ widely in their safety tooling, and the guides call out the ones with real guardrails. Never hand over credentials you cannot revoke, and keep backups of anything an agent is allowed to change.