Complete List of Open Source AI Agents

Multi-Agent Frameworks

LangGraph (MIT) is the production-grade framework from the LangChain ecosystem. It uses a graph-based architecture for precise control over agent execution flow, state management, error recovery, and human-in-the-loop checkpoints. LangSmith integration provides observability for production deployments. Version 0.4, released April 2026, brought improved state persistence. Best for teams that need maximum control over complex agent workflows and are willing to invest in a steeper learning curve.

CrewAI (MIT) provides role-based multi-agent orchestration that maps naturally to how teams work. Define agents with specific roles, goals, and tools, then let them collaborate on tasks. The enterprise tier launched March 2026 adds observability and scheduling. Best for rapid prototyping and team-based agent design where the workflow follows a clear role delegation pattern. The learning curve is significantly lower than LangGraph.

AutoGen (Creative Commons Attribution 4.0) from Microsoft Research enables multi-agent conversations where agents discuss and collaborate to solve problems. It supports group chat patterns, code execution, and human participation in agent conversations. Best for research applications and conversational agent architectures where multiple agents need to debate or refine solutions through dialogue.

CAMEL (Apache 2.0) is a research-focused framework for studying multi-agent communication and cooperation. It provides structured role-playing frameworks and has spawned specialized tools like OWL for research tasks. Best for academic research into agent cooperation patterns and building specialized research agents.

Mastra (MIT) is a TypeScript-first agent framework designed for Next.js, Node, and modern web applications. It provides native integration with frontend and backend JavaScript tooling, making it the best choice for teams building agent-powered web applications in the JavaScript ecosystem. Actively maintained with growing community adoption.

Hermes Agent (Apache 2.0) reached 60,000 GitHub stars in under two months, making it one of the fastest-growing projects in the ecosystem. Its opinionated approach emphasizes simplicity and convention over configuration. Still maturing but the rapid community adoption suggests it fills a real gap for a framework that balances power with simplicity.

Coding Agents

Aider (Apache 2.0) is the strongest terminal-based coding agent. It edits multiple files simultaneously with git integration, creates commits automatically, and includes repository mapping for codebase-aware changes. Supports Claude, GPT-4, Gemini, Llama, Mistral, and dozens of other models. Best for developers who work primarily in a terminal and want reliable multi-file editing with version control.

Cline (Apache 2.0) runs as a VS Code extension providing autonomous coding capabilities with MCP server integration. It supports multi-file reasoning, safety controls for change review, and visual diff display. Best for developers who prefer an IDE-native workflow with safety controls for code changes. Written in TypeScript.

OpenHands (MIT) is the most autonomous coding agent, going beyond editing to full software engineering: repository analysis, implementation planning, code writing, testing, and debugging. Over 70,000 GitHub stars. Runs in a sandboxed environment. Best for well-defined tasks where requirements are clear and the codebase follows standard patterns.

Continue.dev (Apache 2.0) provides AI code assistance in VS Code and JetBrains IDEs with autocomplete, chat, and inline editing. You configure which model handles each function independently, allowing optimization of speed versus quality per task. Best for teams that want flexible AI code assistance across multiple IDEs.

Tabby (Apache 2.0) is a self-hosted code completion service that runs locally on your machine or team server. Supports fine-tuning on your codebase. Best for teams that need a privacy-first alternative to GitHub Copilot that keeps all code on their own infrastructure.

Codex CLI (Apache 2.0) from OpenAI is a terminal coding agent that serves as the reference implementation for many commercial coding tools. Provides a clean interface for code generation and editing through OpenAI models. Tied to the OpenAI ecosystem but fully open source.

Research Agents

OWL (Apache 2.0) ranks first on the GAIA benchmark among open source frameworks. Built on the CAMEL framework, it uses a dual-agent architecture with a planning agent and an execution agent. The planning agent decomposes complex research questions into sub-tasks and the execution agent carries them out. Best for complex, long-horizon research tasks that require dozens of steps.

GPT Researcher (MIT) focuses on end-to-end report generation. Given a research topic, it plans queries, searches the web, gathers information from multiple sources, and produces a structured report with citations. Actively maintained with regular updates. Best for generating comprehensive research reports on well-defined topics with publicly available information.

Perplexica (MIT) provides an open source alternative to commercial AI search engines. Uses multiple agents to search, analyze, and synthesize information from the web with source tracking and citations. Best for teams that want a search-focused research tool rather than a full report generation system.

Browser Automation Agents

Browser Use (MIT) is the leading open source browser automation framework. It combines vision model support, DOM extraction, multi-tab browsing, and parallel agent execution. The LLM has full control over browsing actions rather than following predefined scripts. Best for complex browser tasks that require adaptive navigation and decision-making based on page content.

Stagehand (MIT) from Browserbase provides AI-powered browser automation with a focus on reliability and developer experience. It provides higher-level abstractions for common browser actions while still allowing fine-grained control. Best for developers who want a more structured approach to browser automation with good developer tooling.

Skyvern (AGPL-3.0) automates browser-based workflows using LLMs and computer vision to interact with websites. It focuses on automating business processes that require navigating web applications. Note the AGPL license, which has implications for commercial use. Best for automating repetitive web-based business workflows.

Workflow and Business Automation

n8n (Sustainable Use License / Fair Code) is the most complete workflow automation platform, combining a visual builder with 400+ integrations and native AI capabilities. The AI node supports multiple LLM providers. Self-hosted option gives full data control. Best for teams that need AI-enhanced automation integrated with existing business systems. The visual interface makes it accessible to non-developers.

Dify (Apache 2.0) provides a low-code platform for building AI applications with visual workflow design and built-in RAG capabilities. Supports multiple LLM providers. Web-based interface accessible to non-technical team members. Best for teams that want to build AI-powered tools without extensive coding, particularly when RAG is needed for domain-specific knowledge.

Ontheia (Apache 2.0) is a self-hosted AI agent platform for customer engagement. Features Chain Engine for visual workflow automation, MCP-native tool integration, multi-provider model support, long-term memory via pgvector, RBAC, and GDPR-compliant architecture. Best for customer-facing AI agent deployments where data sovereignty and compliance matter.

Specialized and Emerging Tools

Pydantic AI (MIT) brings structured outputs, validation, and dependency injection to Python agent development. It prioritizes clean application architecture and type safety alongside agent capability. Best for Python developers who value code quality and maintainability and want type-safe agent development.

Composio (Elastic License 2.0) provides a tool integration layer for AI agents, offering pre-built connections to 250+ services with authentication handling. Rather than building an agent framework itself, Composio focuses on making it easy to give any agent access to external tools. Best as a supplementary tool layer for existing agent frameworks.

AgentOps (MIT) provides observability and monitoring specifically designed for AI agents. It tracks agent sessions, LLM calls, tool usage, costs, and errors with a dashboard designed for agent debugging. Best as a monitoring layer added to any agent deployment regardless of framework.

Mem0 (Apache 2.0) adds persistent, intelligent memory to AI agents. It stores, retrieves, and manages long-term conversational context across sessions. Works with multiple agent frameworks. Best for applications where agents need to remember previous interactions, user preferences, or accumulated knowledge across conversation sessions.

How to Use This List

Start by identifying your primary use case: building custom agent workflows (frameworks category), automating code editing (coding agents), conducting research (research agents), automating browser tasks (browser automation), or connecting business systems (workflow automation). Then evaluate the top options in that category based on your specific requirements for model flexibility, self-hosting capability, license compatibility, and community support.

Pay attention to license types because they have fundamentally different implications for commercial use. MIT and Apache 2.0 are the most permissive and allow commercial use with minimal restrictions. AGPL-3.0 requires that any modifications to the code be made available under the same license, which can affect proprietary products. Elastic License 2.0 and similar licenses may restrict SaaS hosting. Review the license details before investing development time in any project.

Check project health beyond GitHub stars. Look at commit frequency in the last 90 days, issue response time, release cadence, and documentation quality. A project with 5,000 stars and weekly releases is a safer bet than one with 50,000 stars and no updates in three months. The open source AI agent ecosystem moves fast, and projects that stop actively developing quickly fall behind.