Agno
Python framework and runtime for building agents, serving them as an API, and managing them in a UI.
Apache-2.0 self-hosted; vendor cloud available
AI Agent Platforms / Open-source agent builders and frameworks
Open-source agent builders and frameworks
Agent builders and frameworks are the tools developers use to create, connect, run, and inspect agents. Some are code libraries, others are visual platforms. This page lists the open-source and source-available options, with the license and hosting model for each. There are 16 in this directory.
What to look for
- Code library versus visual builder, matched to who will maintain the agents
- Model portability, so you are not locked to one provider
- State and memory handling for long-running agents
- Deployment path from prototype to production
- Observability and tracing for debugging agent behavior
- License terms, since some builders separate enterprise features
Platforms
CAMEL
Python framework for building and studying multi-agent systems, tasks, and simulated environments.
Apache-2.0 library for self-managed applications
CrewAI
Python framework for role-based agent crews and event-driven task flows.
MIT self-hosted framework; vendor platform available
Dify
Platform combining a visual workflow builder, RAG pipelines, and agents for LLM applications.
Dify Open Source License (Apache-2.0 based, with conditions) self-hosted; vendor cloud available
Flowise
Visual drag-and-drop builder for AI agents and language-model workflows.
Apache-2.0 core, separately licensed enterprise features self-hosted; vendor cloud available
Haystack
Python orchestration framework for retrieval pipelines, document processing, and agents.
Apache-2.0 library for self-managed applications; vendor cloud available
Langflow
Visual and code-based platform for building agents and workflows served as APIs or MCP servers.
MIT self-hosted; vendor-hosted available
LangGraph
Low-level orchestration framework for long-running, stateful agents with explicit graph control.
MIT library for self-managed applications; vendor platform available
Letta
Framework and server for building stateful agents with persistent memory that learn over time.
Apache-2.0 self-hosted server; vendor-hosted available
LlamaIndex
Python data framework for building agentic LLM applications over documents and data.
MIT library for self-managed applications; vendor cloud available
MetaGPT
Python multi-agent framework that assigns software-company roles to LLMs to complete tasks.
MIT library for self-managed applications
Microsoft Agent Framework
Multi-language framework for building AI agents and multi-agent workflows in .NET and Python.
MIT library for self-managed or cloud applications
OpenAI Agents SDK
Python framework for multi-agent workflows with tools, handoffs, guardrails, and tracing.
MIT library for self-managed applications
Pydantic AI
Typed Python agent framework with broad model support, tools, and durable execution.
MIT library for self-managed applications; vendor platform available
Sim
Collaborative visual workspace for building, deploying, and monitoring agents and workflows.
Apache-2.0 self-hosted; vendor cloud available
smolagents
Compact Python library for agents that act through code snippets or tool calls.
Apache-2.0 library for self-managed applications
Common questions
What is the difference between an agent framework and an agent platform?
A framework is a code library you embed in your own application and control in code. A platform is a runnable product with its own interface, often including a visual builder, deployment, and monitoring. Frameworks give more control; platforms start faster.
Do I need to write code to build an agent?
Not always. Several tools in this category offer visual builders that need little or no code, while others are code-first libraries. The right choice depends on who will build and maintain the agents.
Other categories
Open-source AI coworkers and teammatesOpen-source and self-hostable workflow automation platformsOpen-source browser agentsOpen-source coding agents
See also the guide to choosing a platform and the full landscape and market map.