Open-Source Agent Frameworks

Open-source framework for building and evaluating autonomous data labeling agents. Enables creation of AI agents that can label, process, and improve data quality autonomously with feedback loops and continuous learning capabilities.

Open-source AgentOS project formerly known as AutoGen. AG2 provides multi-agent conversation patterns, swarms, group chats, tool use, and orchestration primitives for building collaborative AI agent systems.

A reliable agent framework built on top of OpenAI Assistants API for creating collaborative swarms of agents with distinct roles and capabilities. Agency Swarm provides customizable agent roles, full control over prompts, and type-safe tools with automatic validation and error correction.

Go framework for building, orchestrating, and deploying LLM-agnostic agentic AI applications. AgenticGoKit supports event-driven multi-agent workflows, MCP tool discovery, and production observability.

Experimental open-source platform for creating self-evolving autonomous AI agents capable of writing and executing their own code. Explores the frontier of autonomous agent development with self-modification and adaptation capabilities for research applications.

Open-source framework for developing autonomous language agents with long-short term memory, tool usage, and multi-agent communication. Features symbolic memory management and comprehensive agent coordination mechanisms for building sophisticated autonomous agent systems.

Developer-centric multi-agent framework that fully embraces asynchronous execution and provides features for building complex, scalable agent applications. Offers distributed agent execution, flexible orchestration, and comprehensive debugging tools for developing production-grade multi-agent systems.

Open-source CLI for scaffolding robust AI agent projects. AgentStack generates project layouts for common agent frameworks and integrates development conventions for tools, observability, and production agent workflows.

Framework designed to facilitate deployment of multiple LLM-based agents in task-solving and simulation applications. Enables building multi-agent environments for software development, social behavior research, and collaborative task completion with flexible agent coordination mechanisms.

Dynamic platform for orchestrating AI agents across various providers with flexible agent and tool management. Features plugin architecture, support for multiple LLM providers, and comprehensive agent lifecycle management for building provider-agnostic agentic applications.

Multi-agent framework, runtime and control plane built for speed, privacy, and scale. Provides primitives for building agents with persistent state, knowledge retrieval, memory management, and best-in-class Model Context Protocol (MCP) support for enterprise-ready agent applications.

Starter kit for building and customizing virtual towns populated with AI agents that can socialize and interact. Built on the Convex platform, provides foundation for creating simulated social environments with autonomous agents for research and entertainment applications.

Apache state-machine framework for applications that make decisions, including agents, simulations, and chatbots. Burr provides graph execution, persistence, monitoring, tracing, and framework-agnostic infrastructure.

Multimodal LLM agent designed to operate smartphone applications by interpreting UI elements and user commands. Enables automation of mobile app interactions through visual understanding and action execution for mobile application testing and automation.

A modular, lightweight framework designed to simplify the creation of multi-agent systems. Atomic Agents focuses on the Input-Process-Output model and atomicity principles, offering developers complete control over agent behaviors and system architecture without unnecessary complexity.

Lightweight open-source framework for building extensible and testable LLM agents with minimal dependencies. Emphasizes simplicity and testability, providing clean abstractions for building production-ready agents without unnecessary complexity or framework lock-in.

Microsoft's open-source framework for creating multi-agent AI applications that can act autonomously or work alongside humans. AutoGen emphasizes conversational agents and multi-agent collaboration, enabling agents to generate, fix, and run code while facilitating cooperation among multiple agents to solve complex tasks.

Open infrastructure for deploying and sharing AI agents without vendor lock-in. BeeAI Agent Stack packages agents as interoperable services with runtime, routing, storage, deployment, and Kubernetes-oriented infrastructure.

Library for multi-agent language game environments used by researchers to study social interaction and collaborative task-solving among LLM agents. Provides standardized environments for evaluating multi-agent communication, cooperation, and competition in conversational settings.

A lean, standalone framework for orchestrating role-playing, autonomous AI agents in collaborative teams. CrewAI allows developers to create specialized agents with distinct roles and goals, optimized for speed and performance while being completely independent of other agent frameworks.

Python framework for building autonomous, resilient, and observable AI agents on Dapr. It combines workflow orchestration, statefulness, telemetry, and security primitives for distributed agent applications.

Open-source framework for building multilingual conversational AI agents and complex dialogue systems. Offers pre-trained models for NER, sentiment analysis, question answering, and dialogue management, with support for 20+ languages and production-ready conversational AI capabilities.

Stanford NLP framework for programming language-model systems with declarative modules instead of hand-written prompts. DSPy supports optimizers for prompts and pipelines used in RAG, agent loops, and compound AI systems.

Pluggable agent SDK and deployment server for agentic applications. Eidolon exposes agents as services and supports componentized LLM, tool, RAG, deployment, and enterprise integration patterns.

Go framework from the CloudWeGo project for building LLM and AI applications. Eino provides component abstractions, graph orchestration, streaming support, and agent development patterns for production systems.

Python framework for coding, building, and evaluating agents with strong support for models, skills, MCP, and ACP. Fast-Agent provides workflow patterns for composing chains, routers, orchestrators, and evaluator loops.

Platform for training, serving, and evaluating large language model-based chatbots with distributed multi-model system support. Features Chatbot Arena for LLM evaluation, model serving infrastructure, and tools for fine-tuning and deploying conversational AI systems.

Python framework for building MCP servers and clients with concise, typed interfaces. FastMCP generates protocol schemas from Python functions and provides a practical toolkit for exposing tools to AI agents.

Original simulation framework from Stanford/Google research demonstrating interactive, human-like social behavior in AI agents. Implements the groundbreaking Generative Agents paper, showcasing believable human-agent interactions in simulated environments for social simulation research.

Go implementation of Google's Agent Development Kit for building, evaluating, and deploying AI agents with a code-first workflow. It brings the ADK agent model to Go teams building cloud-native agent services.

Java implementation of Google's Agent Development Kit for code-first agent development. It provides agent building blocks for JVM teams working with tool use, evaluation, and deployment across Google's agent ecosystem.

TypeScript implementation of Google's Agent Development Kit for building AI agents and multi-agent systems. It offers ADK primitives for Node.js developers and shares documentation with the broader Google ADK project.

Google's open-source Python toolkit for building, evaluating, and deploying AI agents. It supports code-first agent development, multi-agent composition, tool use, and deployment paths through Google's AI and cloud ecosystem.

Modular open-source Python framework for developing AI agent applications with focus on reliability and security. Offers clean abstractions to build agents, systems of agents, pipelines, workflows, and RAG setups with off-prompt business logic definition, memory management, and integration with data sources for production-ready solutions.

An open-source AI orchestration framework by deepset for building production-ready LLM applications, retrieval-augmented generation (RAG) pipelines, and intelligent search systems. Haystack provides modular components for connecting models, vector databases, and file converters into customizable pipelines for working with large document collections.

Declarative platform for building Agent-to-Agent (A2A) native AI agents with focus on inter-agent communication. Enables creation of agent ecosystems where agents communicate and collaborate using declarative configuration and standardized protocols.

Agent library built into the Hugging Face Transformers ecosystem that allows LLMs to use tools and autonomously execute tasks. Provides seamless integration with Hugging Face models and tools, enabling developers to build agentic applications within the Transformers framework with minimal setup.

TypeScript framework for building multi-agent networks with deterministic routing and MCP tooling. AgentKit is designed to run on Inngest's durable execution model for reliable agent workflows.

Foundational Java-based multi-agent framework for building FIPA-compliant systems with ACL messaging, agent discovery, and distributed execution. Provides robust infrastructure for industrial and research agent applications with GUI-based debugging tools and support for distributed agent deployment across networks.

Open-source ChatGPT alternative that operates fully offline on diverse hardware from PCs to multi-GPU clusters. Provides desktop application for running LLMs locally with user-friendly interface, supporting various open-source models for privacy-conscious AI interactions.

JavaScript framework for multi-agent systems using a Kanban board approach for task management and agent coordination. Brings visual task management paradigm to multi-agent orchestration, enabling intuitive workflow design for JavaScript-based agent applications.

A foundational framework for building context-aware reasoning applications powered by large language models. LangChain provides modular components for chaining prompts, integrating external tools, and managing conversational memory, making it suitable for production-grade LLM applications and autonomous agents.

Open-source Java library for building LLM-powered applications on the JVM. LangChain4j provides unified APIs for model providers and vector stores, plus support for tools, agents, RAG, MCP, and enterprise Java frameworks.

A specialized framework within the LangChain ecosystem for building controllable, stateful agents using graph-based execution. LangGraph enables complex multi-step workflows with persistent memory and human-in-the-loop capabilities, trusted by companies like Klarna, Uber, and GitLab.

Java implementation of LangGraph for building agentic architectures in the Java ecosystem. LangGraph4j works with LangChain4j and Spring AI to support graph-based, stateful agent workflows.

Open-source platform for building stateful AI agents with advanced memory. Letta provides long-term memory, agent state management, tool use, and APIs for agents that can learn and improve over time.

Framework for building realtime voice and multimodal AI agents on LiveKit. It provides speech-to-speech pipelines, model integrations, media transport, and production-oriented primitives for interactive voice agents.

A data framework for building LLM applications that specializes in connecting large language models to external data sources. LlamaIndex provides tools for data ingestion, indexing, and retrieval, with over 300 integrations supporting various LLMs, embedding models, and vector stores for RAG applications.

Multi-agent workflow tool with YAML and programmatic support, including human-in-the-loop capabilities. Enables declarative agent workflow definition with configuration-driven approach for building complex multi-agent systems.

Prefect's ambient intelligence Python library for building AI functions, agents, and task-centric workflows. Marvin supports structured generation, classification, extraction, and agent behavior with typed Python interfaces.

TypeScript framework designed for rapid development of AI applications and agentic features. Provides type-safe agent development with modern TypeScript tooling, enabling fast iteration and production deployment of TypeScript-based agentic systems.

Popular Python library for agent-based modeling and simulation, widely used in social science, economics, and supply chain research. Ideal for modeling agent interactions in grid-based or network environments with built-in visualization tools to track agent behaviors and emergent system properties.

A research framework that simulates the structure of a software development team with agents acting as CEO, project manager, and developers. MetaGPT orchestrates these roles to automate software development workflows with minimal human oversight, ideal for autonomous software development pipeline research.

Microsoft framework for building, orchestrating, and deploying AI agents and multi-agent workflows. It supports Python and .NET applications and provides a long-term framework path for Microsoft's agent development stack.

Microsoft's open-source toolkit for developing, testing, evaluating, and deploying LLM applications. PromptFlow uses executable flows that connect prompts, Python code, tools, evaluation, tracing, and deployment workflows.

Minimal Python and TypeScript library for building LLM applications with typed calls, structured outputs, tool calling, streaming, async support, and MCP. Mirascope focuses on lightweight abstractions over provider APIs.

Agent framework connecting ModelScope's open-source models to real-world applications. Provides integration layer for leveraging ModelScope's model ecosystem in agentic applications with pre-built tools and model connectors.

Python framework for orchestrating AI agents and managing complex LLM-driven tasks. Nexus supports multi-agent coordination, structured workflows, persistent state, and tool integration for research and application development.

NVIDIA open-source library for connecting and optimizing teams of AI agents. The toolkit integrates with the NeMo ecosystem and supports enterprise multi-agent workflows that use NVIDIA AI infrastructure.

Popular tool for running large language models (Llama 3, Mistral, Gemma, etc.) locally with simplified setup and customization. Supports macOS, Windows, and Linux with easy model management, REST API, and integration with development tools for local LLM deployment.

Open-source coding agent that enables language models to execute code (Python, JavaScript, Shell, etc.) locally on a user's computer. Provides natural language interface for code execution, file manipulation, and system operations, functioning as a local alternative to ChatGPT's Code Interpreter.

An open research platform for language agents that aims to democratize agent development for researchers and practitioners. OpenAgents provides a comprehensive framework for building and evaluating language agents with multiple capabilities and interaction patterns.

Framework that combines LLMs with domain-specific expert models to solve advanced tasks through collaborative intelligence. Enables agents to leverage specialized models for complex problem-solving, bridging general language understanding with domain-specific expertise for enhanced task performance.

Official JavaScript and TypeScript framework from OpenAI for multi-agent workflows and voice agents. It provides agent orchestration primitives, tool use, handoffs, guardrails, tracing, and MCP integration for Node.js applications.

Official Python framework from OpenAI for building multi-agent workflows with agents, handoffs, guardrails, sessions, tracing, and realtime voice support. The SDK is lightweight and designed for production agent applications.

Lightweight, experimental open-source framework from OpenAI for multi-agent orchestration and coordination. Explores agent routines and handoffs for simple multi-agent coordination patterns using OpenAI's Chat Completions API, designed for educational purposes and research into agent interaction patterns.

Python library for cognitive architectures and multi-agent teams. Orchestra provides abstractions for coordinating specialized agents with shared memory, task delegation, and composable orchestration patterns.

Open-source high-performance execution engine for agentic workflows from Harvard CNS. Orla focuses on scheduling, parallel execution, and efficient runtime support for complex agent task graphs.

Structured-output library for reliable language model generation. Outlines constrains model outputs with types, schemas, and generation controls used in tool-calling, data extraction, and agent workflows.

Python-based framework for multi-agent system development, execution, and management with 100% Python implementation. Features network-based communication for distributed computation, commonly used in academic and industrial IoT applications for building distributed multi-agent systems.

Framework for building AI agents with memory, tools, and production scalability. Enables developers to create autonomous assistants using Pythonic object-oriented principles, featuring structured outputs, knowledge bases, and seamless integration with popular LLM providers for building production-ready agentic systems.

Open-source Python framework for voice and multimodal conversational AI. Pipecat provides composable real-time pipelines for speech, language models, media transport, and interactive voice agent applications.

Minimal LLM framework built around a small graph abstraction for agents, workflows, and RAG applications. Pocket Flow emphasizes a tiny dependency surface and lets agents build or modify agent workflows.

Open-source tool for querying documents using LLMs in a secure, offline-capable environment ensuring complete data privacy. Enables users to interact with their documents using AI without internet connectivity, keeping all data local and secure for privacy-sensitive applications.

A Python agent framework that brings Pydantic's famous type safety and ergonomic developer experience to AI agent development. Pydantic AI focuses on providing type-safe, well-structured agent development with automatic validation and error handling.

Official agent framework and application scaffold for Alibaba's Qwen model family. Qwen-Agent supports function calling, MCP tools, code interpreter workflows, RAG, browser extension use cases, and multi-agent applications.

Open-source conversational AI framework with robust NLU and dialogue management capabilities, trusted by enterprises for building production-ready conversational agents. With over 50 million downloads, Rasa provides tools for building on-premise, production-ready conversational agents across text and voice with full control over data and infrastructure.

Python framework for distributed computing that excels in multi-agent reinforcement learning and real-time decision systems. Scales agents across clusters using actor model for parallel execution, critical for training and deploying sophisticated multi-agent reinforcement learning applications in production environments.

LLM-based agent designed to control real-world applications by interacting with RESTful APIs. Enables agents to interact with web services and APIs for practical task execution, bridging LLM capabilities with real-world application integration.

Rust framework for building modular LLM applications and agentic workflows. Rig provides provider integrations, vector-store support, and scalable abstractions for Rust-native agent and RAG applications.

Microsoft's open-source SDK for integrating AI Large Language Models with conventional programming languages including C#, Python, and Java. Semantic Kernel serves as middleware that enables rapid delivery of enterprise-grade AI solutions with built-in planning capabilities and seamless integration with existing business systems.

A Hugging Face framework for building AI agents that integrates seamlessly with the Hugging Face ecosystem. Smolagents provides tools for creating intelligent agents capable of data retrieval, summarization, and code execution, with strong community engagement and regular updates.

Spring application framework for AI engineering. Spring AI integrates model providers, vector stores, embeddings, tool calling, RAG pipelines, and agentic patterns into the Spring and Spring Boot ecosystem.

AWS's toolkit for building AI agents that integrate with Amazon Bedrock and other AWS services. The SDK focuses on production readiness with first-class OpenTelemetry tracing and native AWS integrations, designed for enterprise-grade agent development in cloud environments.

Open-source development platform for AI Agents offering Python SDK, cloud deployment, serverless hosting, and vector search. Provides complete infrastructure for building, deploying, and scaling AI agents with managed services and developer-friendly APIs.

Enterprise-grade open-source framework designed for scaling hierarchical swarms of AI agents in production environments. Provides tools for building and managing large-scale multi-agent systems with hierarchical coordination, task distribution, and centralized monitoring for production deployments.

Rust framework for fast streaming indexing, querying, RAG, and agentic LLM applications. Swiftide uses async Rust patterns for high-performance LLM pipelines and agent workflows.

Open-source, self-hosted AI coding assistant similar to GitHub Copilot with self-contained setup and OpenAPI integration. Provides code completion, RAG-based code chat, and IDE integration for building private coding assistants with full data control.

Research framework for optimizing LLM systems using textual feedback as gradients. TextGrad applies natural-language critique to improve prompts, model pipelines, and agent behaviors through iterative optimization.

Python framework from the tRPC group for agent building, orchestration, tool integration, session and long-term memory, service deployment, and observability. It provides a foundation for reliable and extensible agent applications.

Reliable agent framework featuring Model Context Protocol (MCP) support and verification layers for robust execution. Emphasizes reliability and correctness in agent operations with built-in verification mechanisms and error handling.

Open-source TypeScript AI agent framework and engineering platform for building agent applications. VoltAgent includes primitives for agents, workflows, memory, tool use, MCP integration, and an optional cloud console.