A curated list of awesome LLM frameworks and agent development tools. If you have a suggestion, feel free to open an issue or pull request. (Last updated: 2026-03-22)
-
CrewAI - Framework for orchestrating role-playing AI agents
46,791 stars · 6,328 forks · 283 contributors · 437 issues · Python · MIT
- Role-based agent design
- Multi-agent collaboration
- Flexible memory system
- Built-in error handling
-
Langchain - Building applications with LLMs through composability
130,508 stars · 21,500 forks · 470 contributors · 488 issues · Python · MIT
- Modular and extensible architecture
- Unified interface for LLMs
- Pre-built agent toolkits
- CSV, JSON, and SQL agents
- Python and Pandas integration
- Vector store capabilities
-
Microsoft AutoGen - Framework for building multi-agent conversational systems
55,980 stars · 8,425 forks · 445 contributors · 690 issues · Python · CC-BY-4.0
- Multi-agent architecture
- Customizable agents
- Code execution support
- Flexible human involvement
- Advanced conversation management
-
OpenManus - Open-source AI agent platform for building general-purpose agents
55,409 stars · 9,685 forks · 53 contributors · 528 issues · Python · MIT
- Modular architecture for customization
- Data analysis and visualization agents
- Reinforcement learning integration (OpenManus-RL)
- No invitation code required
- Built by MetaGPT contributors
-
Llama Index - Data framework for LLM applications
47,844 stars · 7,060 forks · 475 contributors · 262 issues · Python · MIT
- Advanced indexing and retrieval
- Support for 160+ data sources
- Customizable RAG workflows
- Structured data handling
- Query optimization
-
Microsoft Semantic Kernel - Integration framework for AI models
27,522 stars · 4,516 forks · 393 contributors · 501 issues · C# · MIT
- Enterprise-grade security
- Multi-language support
- Plugin architecture
- Responsible AI features
- Memory management
-
Dify - Open-source framework for LLM applications
133,884 stars · 20,854 forks · 462 contributors · 808 issues · TypeScript · NOASSERTION
- Visual prompt orchestration
- Long context integration
- API-based development
- Multi-model support
- RAG pipeline
-
Haystack - End-to-end NLP framework
24,582 stars · 2,670 forks · 326 contributors · 100 issues · MDX · Apache-2.0
- Document processing
- Neural search
- Question answering
- Semantic search
- Agent capabilities
-
Embedchain - Framework for ChatGPT-like bots
50,633 stars · 5,650 forks · 269 contributors · 589 issues · Python · Apache-2.0
- Multi-source data ingestion
- Automated embedding
- Context window management
- Multiple LLM support
- RAG optimization
-
Google ADK - Agent Development Kit for building, evaluating, and deploying AI agents
18,517 stars · 3,111 forks · 254 contributors · 629 issues · Python · Apache-2.0
- Code-first development approach
- Modular multi-agent systems
- Model-agnostic and deployment-agnostic
- Built-in developer UI (adk-web)
- Deploy on Cloud Run or Vertex AI
-
SuperAGI - Open-source autonomous AI agent framework
17,287 stars · 2,182 forks · 62 contributors · 221 issues · Python · MIT
- Customizable agent workflows
- Tool creation framework
- Performance monitoring
- Resource management
- Multi-vector memory storage
-
Kiln AI - Tools for building AI products including agents, evals, RAG systems, and fine-tuning
4,712 stars · 349 forks · 12 contributors · 42 issues · Python · NOASSERTION
- Free desktop application for no-code development
- Evaluation frameworks
- RAG system building
- Synthetic data generation
- Agent development tools
-
AGiXT - Scalable framework for AI agents
3,166 stars · 442 forks · 41 contributors · 4 issues · Python · MIT
- Multi-provider support
- Chain of thought processing
- Extensible plugin system
- Command chaining
- Web UI included
-
XAgent - Autonomous LLM-based agent framework
8,516 stars · 896 forks · 33 contributors · 57 issues · Python · Apache-2.0
- Human-like planning
- Autonomous task decomposition
- Tool learning capabilities
- Advanced error recovery
- Built-in action validation
-
Neurolink - Multi-provider AI agent framework with workflow orchestration capabilities, unifying 12+ providers (OpenAI, Google, Anthropic, AWS, Azure, Groq, Together AI, Mistral, Cohere, Fireworks, Cloudflare, Ollama)
119 stars · 95 forks · 53 contributors · 424 issues · TypeScript · MIT
- Multi-agent framework with workflow orchestration
- Unified interface for 12+ AI providers
- Edge-first architecture with local/cloud deployment
- Production-grade streaming and tool calling
- Battle-tested at Juspay (15M+ requests/month)
-
OpenAgents - Open platform for language agents
4,729 stars · 523 forks · 16 contributors · 15 issues · Python · Apache-2.0
- Data analysis capabilities
- Web browsing integration
- Coding assistance
- Plugin ecosystem
- Interactive visualization
-
AI Legion - Swarm framework for autonomous agents
1,429 stars · 174 forks · 6 contributors · 9 issues · TypeScript · MIT
- Multi-agent coordination
- Dynamic task allocation
- Emergent behavior support
- Flexible agent roles
- Real-time collaboration
-
Agent Protocol - Unified interface for AI agents
1,455 stars · 178 forks · 14 contributors · 43 issues · Python · MIT
- Standardized communication
- Language-agnostic design
- Tool integration specs
- Interoperability focus
- Protocol versioning
-
Agents.js - JavaScript framework for building AI agents
- Browser-native implementation
- Event-driven architecture
- Tool abstraction layer
- Memory management
- Real-time processing- Cache-to-Cache - Novel paradigm for direct semantic communication between LLMs via KV-Cache
87 stars · 6 forks · 2 contributors · 2 issues · Python · Apache-2.0
- Direct semantic communication via KV-Cache
- Multi-agent system integration
- Neural cache projection and fusion- CAMEL - Communicative Agents for "Mind" Exploration
14,875 stars · 1,638 forks · 171 contributors · 599 issues · Python · Apache-2.0
-
Role-playing framework
-
Task-oriented dialogue
-
Multi-agent conversations
-
Behavioral analysis
-
Cognitive architecture- BabyAGI - Lightweight framework for AI task management
-
Task prioritization
-
Autonomous execution
-
Memory persistence
-
Goal-oriented planning
-
Resource optimization- Autonomous-GPT - Framework for autonomous GPT-4 agents
179,412 stars · 46,089 forks · 433 contributors · 273 issues · Python · NOASSERTION
- Internet access capabilities
- Long-term memory
- Goal-oriented behavior
- File operations
- Command execution- MetaGPT - Multi-agent framework for software development
59,615 stars · 7,299 forks · 115 contributors · 57 issues · Python · MIT
- Role-based development
- Code generation
- Project management
- Documentation writing
- Testing automation- GenoMAS - Multi-agent framework for scientific discovery and automated data analysis
123 stars · 19 forks · 1 contributors · 0 issues · Python · MIT
-
Guided planning framework
-
Typed message-passing protocol
-
Heterogeneous LLM architecture
-
Domain-agnostic design
-
Scientific workflow automation- minions - Extensible framework for AI assistants
-
Custom behavior definition
-
Tool integration
-
State management
-
Event handling
-
Parallel execution- ix - Autonomous agent framework
1,037 stars · 129 forks · 5 contributors · 14 issues · Python · MIT
- Visual workflow builder
- Sandbox environments
- Tool integration
- Process monitoring
- Agent collaboration- saplings - Build smarter agents using tree search
269 stars · 17 forks · 4 contributors · 0 issues · Python · Apache-2.0
- Boost reasoning abilities
- Supports popular search algorithms
- Minimal setup, 2 lines of code- Smolagents - Minimalist framework for building powerful agents
24,406 stars · 2,180 forks · 189 contributors · 327 issues · Python · Apache-2.0
- Code-first approach
- Multi-agent orchestration
- LLM provider flexibility
- Tool integration
- Hub integration for sharing- Flowise - Drag & drop UI framework for building LLM flows
47,784 stars · 23,485 forks · 268 contributors · 747 issues · TypeScript · NOASSERTION
- Visual flow builder
- Custom LLM integrations
- API generation
- Authentication support
- Docker deployment- Pydantic AI - Production-grade agent framework built on Pydantic
14,433 stars · 1,568 forks · 348 contributors · 478 issues · Python · MIT
- Type-safe development
- Multi-model support
- Structured responses
- Dependency injection
- Logfire integration- Upsonic - Reliable agent framework that support MCP.
7,778 stars · 717 forks · 35 contributors · 8 issues · Python · MIT
- Easy-to-activate reliability layers
- Model Context Protocol (MCP)
- Integrated Browser Use and Computer Use
- Isolated environment to run agents
- Task-Centric Design- EvoAgentX - Building a Self-Evolving Ecosystem of AI Agents
2,608 stars · 217 forks · 24 contributors · 13 issues · Python · NOASSERTION
- Easy Agent and Workflow Customization
- Workflow Optimization & Self-Evolving
- Agent/workflow evolution algorithms integrated
- Execution Toolkit
-
Portia AI - Open source framework for predictable, controllable and authenticated agents.
1,141 stars · 101 forks · 26 contributors · 44 issues · Python · NOASSERTION
- Structured planning
- Stateful execution
- Human in the loop controls
- Tool catalogue with built-in auth
- MCP support- Agentic Radar - A security scanner for agentic workflows
848 stars · 106 forks · 8 contributors · 10 issues · Python · Apache-2.0
- Scans agentic workflow source code
- Finds vulnerabilities (CVE & OWASP)
- Generates interactive reports
- Suggests remediation steps
- Supports popular agentic workflows- AgentFlow - Trainable multi-agent framework with in-the-flow optimization
1,477 stars · 190 forks · 4 contributors · 7 issues · Python · MIT
- Four specialized modules (planner, executor, verifier, generator)
- Flow-GRPO reinforcement learning
- Tool integration (math, coding, scientific, search)
- Stanford research project
- Performance gains over monolithic approaches- Mastra - TypeScript AI agent framework with assistants, RAG, and observability
20,636 stars · 1,483 forks · 297 contributors · 366 issues · TypeScript · NOASSERTION
- Type-safe development
- Multi-model support, and router (GPT-4, Claude, Gemini, Llama)
- Structured responses
- API generation
- MCP generation
- Tool integration- Flappy - Production-ready LLM agent SDK
307 stars · 23 forks · 7 contributors · 10 issues · Rust · Apache-2.0
- Production-grade reliability
- Type-safe agent development
- Multi-language support
- Performance optimized
- Developer-friendly SDK- CleverBee - Deep research assistant agent with web browsing capabilities
309 stars · 16 forks · 2 contributors · 1 issues · Python · AGPL-3.0
- Interactive web UI via Chainlit
- MCP tool support for external integrations
- Multi-LLM research (configurable LLMs)
- Automated web browsing with Playwright
- Token tracking and cost estimation
-
RAI - Agentic framework for robotics using ROS 2
472 stars · 61 forks · 24 contributors · 61 issues · Python · Apache-2.0
- ROS 2 integration
- Complex action execution
- Scenario-based testing
- Voice interaction support
- Vendor-agnostic architecture
-
Floom - AI gateway and agent/pipeline marketplace
45 stars · 4 forks · 2 contributors · 0 issues · C# · MIT
- AI pipeline orchestration
- Agent marketplace integration
- Kubernetes-style architecture
- Streamlined AI integration
- DevOps-friendly deployment
-
CoreAgent - Minimalist agent framework with stateful tools
25 stars · 3 forks · 2 issues · Python
- Simplicity-first design
- Stateful tool support
- Multi-agent coordination
- Shared state management
- Built-in tools library- Project Alice - Framework and platform for building and deploying agentic workflows
253 stars · 34 forks · 1 contributors · 1 issues · Python · BSD-3-Clause
- No-code and coding interface
- Visual workflow builder
- Live demo available
- Workflow deployment platform
- Alpha stage development- AgentSquare - Automatic LLM agent search in modular design space
199 stars · 14 forks · 10 issues · HTML
- Module evolution and recombination
- Planning, Reasoning, Tool Use, Memory modules
- Adaptive agent search framework
- Unified framework for six agent tasks
- Tsinghua University research project- Flock - Declarative multi-agent LLM orchestration using blackboard architecture and typed contracts
87 stars · 9 forks · 9 contributors · 15 issues · Python · MIT
- Blackboard coordination pattern
- Easily orchestrate complex patterns like multi-output fan-out
- Fluent DX - write complex flows in a few lines of code
- Agents adhere to strict contracts instead of brittle natural language prompts
- Production-grade observability and zero-trust security model
-
hcom - Let AI agents message, watch, and spawn each other across terminals
156 stars · 19 forks · 2 contributors · 3 issues · Rust · MIT
- Works with Claude Code, Gemini CLI, Codex CLI, and OpenCode
- Agents message each other mid-turn, detect file edit collisions, read transcripts
- Agents view terminal screens, subscribe to activity, spawn/fork/resume each other
- TUI dashboard, cross-device relay via MQTT, Python API
- Multi-agent workflow scripts: debates, ensemble refinement, code review watchers
- Just prefix
hcomin front of your existing tool command
-
everyrow - AI-powered data operations SDK for running LLM agents on pandas DataFrames
18 stars · 3 forks · 12 contributors · 4 issues · Python · MIT
- Screen, rank, dedupe, merge rows with natural language
- Web research agents applied per row
- Scales to tens of thousands of rows
- Claude Code integration
-
Axar - Minimal TypeScript agentic framework for building production-ready LLM applications
158 stars · 14 forks · 7 contributors · 4 issues · TypeScript · Apache-2.0
- Strongly-typed agent architecture with first-class TypeScript support
- Decorator-based API for defining agents, tools, and workflows
- Built-in structured I/O and validation (Zod-powered)
- Lightweight, production-oriented design
- Easy integration with OpenAI, Anthropic, Gemini, and other models
-
PraisonAI - Production-ready Multi-AI Agents framework with self-reflection
5,710 stars · 781 forks · 29 contributors · 64 issues · Python · MIT
- 100+ LLM support via LiteLLM
- MCP Protocol integration
- Agentic workflows (route, parallel, loop, repeat)
- Built-in memory (short-term, long-term, entity)
- Self-reflection for improved responses
- Both Python and JavaScript SDKs
-
Tambo - React framework for building AI-powered applications with generative UI and MCP support
11,098 stars · 556 forks · 56 contributors · 32 issues · TypeScript · MIT
- Component library for AI-driven UI composition
- MCP (Model Context Protocol) integration
- TypeScript-first with full type safety
- SSR compatible (Next.js, Remix, etc.)
-
OpenAgents - Open-source platform for building AI agent networks with multi-protocol communication and orchestration
1,774 stars · 219 forks · 15 contributors · 55 issues · Python · Apache-2.0
- Multi-protocol support: WebSocket, gRPC, HTTP, MCP, A2A
- Multi-agent orchestration with centralized and decentralized topologies
- Compatible with Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocol
- Python SDK available on PyPI
-
Cordum - Safety-first agent orchestration platform with pre-dispatch policy evaluation and MCP server support
458 stars · 19 forks · 4 contributors · 15 issues · Go · NOASSERTION
- Pre-dispatch safety policy evaluation
- Output scanning and quarantine
- Job scheduling and workflow engine
- MCP server integration
- gRPC safety kernel
-
AgentField - Open-source infrastructure for AI backends with cryptographic identity
1,055 stars · 158 forks · 13 contributors · 37 issues · Go · Apache-2.0
- W3C DIDs give every agent cryptographic identity
- Guided autonomy: agents reason freely within policy boundaries
- Async-native execution for workflows running minutes to days
- Cross-agent discovery and RPC without pre-federation
- Built-in memory fabric with vector search
-
DeepAnalyze - Agentic LLM for autonomous data science
3,839 stars · 550 forks · 12 contributors · 19 issues · Python · MIT
- Agentic LLM without any predefined workflow
- Autonomous orchestration and adaptive optimization
- Curriculum-based agentic training in real-world environments
- Supporting data tasks: data preparation, analysis, modeling, visualization, and reporting
- Supporting data research: deep research across unstructured, semi-structured, and structured data
-
agent-opt - Open-source optimization engine for iterative prompt refinement and agent workflow performance
51 stars · 3 forks · 3 contributors · 0 issues · Python · NOASSERTION
- Six optimization algorithms: Random, Bayesian, ProTeGi, Meta-Prompt, PromptWizard, GEPA
- Flexible evaluation via heuristic metrics and LLM-as-a-judge
- Works with any LLM provider through LiteLLM
- Clean abstraction layer for custom optimizers and evaluators
- Built-in logging, progress tracking, and reproducible experimentation- Agent OS - Safety-first kernel for governing autonomous AI agents with POSIX-inspired primitives
68 stars · 20 forks · 14 contributors · 0 issues · Python · MIT
- Policy engine with violation guarantees
- Resource quotas and audit logging
- Integrations with CrewAI, LangChain, AutoGen, Semantic Kernel
- MCP server support
-
AgentMesh - Secure trust layer for multi-agent ecosystems with zero-trust governance
17 stars · 6 forks · 4 contributors · 0 issues · Python · MIT
- Ed25519 cryptographic agent identity
- Delegation chains with bounded depth
- HTTP trust middleware for Flask/FastAPI
- A2A protocol compatible
-
Quorum - Multi-agent AI discussion system for structured debates
80 stars · 10 forks · 1 contributors · 0 issues · Python · NOASSERTION
- 7 discussion methods: Standard, Oxford, Socratic, Delphi, Brainstorm, Tradeoff, Advocate
- Multi-phase consensus: independent answers, critique, discussion, synthesis
- Supports Claude, GPT, Gemini, Grok, and local Ollama models
- Terminal UI with real-time streaming
- Auto-discovery of local Ollama models
-
auto-co - Autonomous AI company OS with 14 specialized agents that run a startup end-to-end
22 stars · 3 forks · 1 contributors · 0 issues · TypeScript · MIT
- 14 expert-persona agents: CEO (Bezos), CTO (Vogels), Critic (Munger), CFO, marketer, engineer, QA, DevOps, and more
- Continuous bash loop — agents debate, decide, and ship real deployments autonomously
- Shared markdown consensus file as the cross-cycle relay baton
- Human escalation via Telegram for true blockers only
- Repo is a live company: built its own landing page, Docker stack, and monitoring across 13 autonomous cycles