AI Agent Frameworks, Builders, and Runtimes
23 tools · listed in dataset order, no ranking · 1 discontinued, listed last
Tools for building your own agents, in four shapes: code-first frameworks you program against (LangGraph, CrewAI, Microsoft Agent Framework, Google ADK, Mastra), visual builders where agents are assembled without code (Dify, AutoGPT, ZooWork), runtimes that run and serve agents (Docker Agent, TrueForge, Agno's AgentOS), and control planes that supervise agents you already use (Paperclip). Pick the shape first, then check language support and whether the project is still actively maintained: AutoGen and Semantic Kernel are superseded by Microsoft Agent Framework, and Flowise is discontinued.
LangGraph
LangChain
Graph-Based Framework for Stateful Multi-Agent Workflows
Best for: Developers building production multi-agent systems that need fine-grained control over state, execution flow, and human-in-the-loop checkpoints — and who are willing to trade setup time for that control
CrewAI
CrewAI
Multi-Agent Orchestration Platform
Best for: Orchestrating autonomous agent teams for enterprise tasks
AutoGen
Microsoft
Conversational Multi-Agent Framework by Microsoft
Best for: Existing AutoGen users maintaining projects, and developers experimenting with conversational multi-agent patterns or building iterative workflows (code generation + review, research + verification) where the solution emerges from agent dialogue rather than a predefined execution graph
Semantic Kernel
Microsoft
Model-Agnostic AI Agent SDK
Best for: Maintaining or extending existing Semantic Kernel apps in .NET, Python, and Java
LlamaIndex
LlamaIndex
Data Framework for LLM Apps and RAG Pipelines
Best for: Developers building RAG systems and document-grounded agents who need intelligent data parsing, indexing, and retrieval — especially teams with large or complex document sets
Pydantic AI
Pydantic
Type-Safe Python Agent Framework
Best for: Type-safe Python agents with validated structured outputs
Smolagents
Hugging Face
Minimal Code-First Agent Library
Best for: Minimal code-first agents with Hugging Face ecosystem
Dify
LangGenius
No-Code Agentic Workflow Platform
Best for: No-code agentic workflows and RAG pipelines
Mastra
Mastra
TypeScript AI Agent Framework
Best for: TypeScript-first AI agents and workflows for Node.js teams
VoltAgent
VoltAgent
TypeScript AI Agent Engineering Platform
Best for: TypeScript multi-agent apps with built-in observability
Microsoft Agent Framework
Microsoft
Multi-Agent Orchestration Framework
Best for: Python and .NET multi-agent orchestration, especially with Azure and Microsoft Foundry
IntentKit
Crestal Network
Cloud-Native Agent Cluster Framework
Best for: Cloud-native collaborative AI agent cluster deployment
Agent Zero
Agent Zero
Open-Source Autonomous Agent Framework
Best for: Developers and tinkerers who want a transparent, self-hosted agentic framework they can extend and connect to any model
TrueForge
TrueFoundry
Open-Source Runtime for Production AI Agents
Best for: Platform and application teams that want a model-neutral, self-hosted runtime for reusable production agents
Paperclip
Paperclip Labs, Inc.
Open-Source Control Plane for Running Teams of AI Agents
Best for: Technical founders, builders, and small teams already running several coding or general-purpose agents who want budgets, approvals, and an audit trail around them
AutoGPT
Significant Gravitas
Agent Platform With Chat-Built Agents, a Visual Builder, and Self-Hosting
Best for: Individuals and small teams who want to automate recurring research, content, and operations work with agents built through chat or a visual builder, with the option to self-host
Google ADK
Google Agent Development Kit (ADK): Open-Source Multi-Language Agent Framework
Best for: Developers building production agents in Python, TypeScript, Go, Java, or Kotlin who want an open-source, code-first framework with first-class Gemini support and a direct path to Google Cloud
TradingAgents
Tauric Research
TradingAgents: Open-Source Multi-Agent LLM Framework for Trading Research
Best for: Researchers, students, and technically minded investors who want an open-source, extensible multi-agent framework for studying LLM-driven market analysis and backtesting it
ZooWork
Serendipity One
ZooWork: No-Code Agent Builder and Managed Agent API for Delivering AI Agents to Teams and Clients
Best for: Consultants, agencies, forward-deployed engineers, and operations teams who want to package a repeatable workflow as an agent with approvals and schedules, deliver it to several teams or clients, and optionally embed it in their own product through an API
Agno
Agno Inc.
Agno: Open-Source Python Agent Framework and Self-Hosted AgentOS Runtime
Best for: Python teams that want one open-source stack to build agents, teams and workflows and run them in their own cloud as an API, MCP server and chat-app bot, with data kept in their own database and an optional hosted UI for operating them
Docker Agent
Docker
Docker Agent (formerly cagent): Open-Source Runtime for YAML-Defined AI Agent Teams
Best for: Developers and platform teams already using Docker who want versionable, shareable agents and multi-agent teams defined in config, with MCP tools and a free choice of cloud or local models
Agent S
Simular
Agent S: Open-Source Computer-Use Agent Framework That Operates macOS, Windows, and Linux Desktops
Best for: Developers and AI researchers who want an open-source, model-agnostic computer-use agent they can run on their own macOS, Windows, or Linux machine, inspect, and modify, and who are comfortable hosting a grounding model and paying model APIs directly
Flowise
Workday (FlowiseAI)
Visual Drag-and-Drop Agent Builder (Sunset)
Best for: Existing Flowise users planning a migration, or teams willing to maintain their own fork
How to choose a framework, builder, or runtime
Pick the shape: framework, builder, runtime, or control plane
Code-first frameworks such as LangGraph, CrewAI, Pydantic AI, Mastra, and Google ADK give developers full control and live in your codebase. Visual builders such as Dify and AutoGPT let people assemble agents on a canvas or from blocks, and ZooWork builds them from a chat description. Runtimes run and serve agents: Docker Agent defines agent teams in a YAML or HCL file, TrueForge adds persistent sessions, sandboxed tools, and approvals, and Agno pairs its framework with AgentOS. Paperclip is different again: it supervises agents you already use, such as Claude Code and Codex, with budgets and approval gates. These shapes often combine rather than compete.
Match your language, then check maintenance
LangGraph supports Python and JavaScript/TypeScript. CrewAI, Pydantic AI, Smolagents, Agno, and LlamaIndex are Python (LlamaIndex's TypeScript version is deprecated). Mastra and VoltAgent are TypeScript. Microsoft Agent Framework supports Python and .NET, with Go in preview, and Google ADK covers Python, TypeScript, Go, Java, and Kotlin. Before committing, check the project's status: AutoGen is in maintenance mode, Semantic Kernel receives fixes while Microsoft steers new work to Agent Framework, and Flowise's repository was archived in August 2026.
Orchestration model matters more than feature count
LangGraph uses explicit graphs with checkpoints for state and human review. CrewAI organizes agents into role-based crews. Smolagents has agents act by writing code. Docker Agent declares a team in configuration instead of code. These are different ways of coordinating agents, so prototype your actual workflow in the model that matches it before comparing feature lists.
Languages, maintenance status, and license for every listing
Checked against each project's official repository, release feed, and docs on October 15, 2026. Active means a release or commit within the last few weeks. The last column is our editorial note on fit.
- Shape
- Code-first framework (graphs)
- Languages
- Python, JavaScript/TypeScript
- Status (October 2026)
- Active
- License
- MIT
- Good fit for
- Explicit control over state, branching, and human-in-the-loop checkpoints
- Shape
- Code-first framework; visual Crew Studio in the paid AMP platform
- Languages
- Python
- Status (October 2026)
- Active
- License
- MIT
- Good fit for
- Role-based agent teams that are quick to prototype
- Shape
- Code-first framework
- Languages
- Python, .NET; Go in preview
- Status (October 2026)
- Active; 1.0 since April 2026
- License
- MIT
- Good fit for
- New Microsoft-stack projects; successor to AutoGen and Semantic Kernel
- Shape
- Code-first framework; experimental visual builder (Python)
- Languages
- Python, TypeScript, Go, Java, Kotlin
- Status (October 2026)
- Active
- License
- Apache-2.0
- Good fit for
- Teams working in several languages; optimized for Gemini, runs other models
- Shape
- Code-first framework
- Languages
- Python
- Status (October 2026)
- Active
- License
- MIT
- Good fit for
- Typed tools and validated, structured outputs
- Shape
- Minimal code-first library
- Languages
- Python
- Status (October 2026)
- Maintained; last release May 2026
- License
- Apache-2.0
- Good fit for
- Small agents that act by writing code
- Shape
- Code-first data and RAG framework
- Languages
- Python (the TypeScript version is deprecated)
- Status (October 2026)
- Active; the company's focus has shifted to LlamaParse
- License
- MIT
- Good fit for
- Agents grounded in documents and other data
- Shape
- Code-first framework plus AgentOS runtime
- Languages
- Python
- Status (October 2026)
- Active
- License
- Apache-2.0
- Good fit for
- Python teams serving agents from their own cloud
- Shape
- Code-first framework; Agent Builder for stored agents
- Languages
- TypeScript
- Status (October 2026)
- Active
- License
- Apache-2.0 core; enterprise features under a separate license
- Good fit for
- TypeScript and Node.js applications
- Shape
- Code-first framework plus VoltOps console
- Languages
- TypeScript
- Status (October 2026)
- Active
- License
- MIT
- Good fit for
- TypeScript multi-agent systems with built-in observability
- Shape
- Code-first framework
- Languages
- Python, .NET
- Status (October 2026)
- Maintenance mode: no new features; Microsoft points new projects to Agent Framework
- License
- MIT
- Good fit for
- Existing AutoGen projects
- Shape
- Code-first SDK
- Languages
- C#/.NET, Python, Java
- Status (October 2026)
- Superseded by Microsoft Agent Framework; v1.x gets critical fixes and security patches
- License
- MIT
- Good fit for
- Existing Semantic Kernel applications
- Shape
- Visual workflow builder
- Languages
- No code needed; self-host or cloud
- Status (October 2026)
- Active
- License
- Source-available (Apache 2.0 with added conditions)
- Good fit for
- RAG apps and workflows built on a canvas
- Shape
- Visual block builder and chat-built agents
- Languages
- No code needed; self-host or cloud
- Status (October 2026)
- Active; platform in beta
- License
- Platform source-available (PolyForm Shield)
- Good fit for
- Recurring automations on schedules and webhooks
- Shape
- Hosted chat-based agent builder plus managed agent API
- Languages
- No code needed; TypeScript and Python SDKs
- Status (October 2026)
- Active; Managed Agent API in developer preview
- License
- Proprietary
- Good fit for
- Delivering agents with approval gates to teams and clients
- Shape
- Visual drag-and-drop builder
- Languages
- No code needed
- Status (October 2026)
- Discontinued; repository archived August 2026
- License
- Apache-2.0
- Good fit for
- Not for new projects
- Shape
- Config-first runtime
- Languages
- YAML or HCL files; Go SDK for embedding
- Status (October 2026)
- Active
- License
- Apache-2.0
- Good fit for
- Repeatable agent teams shared through registries like container images
- Shape
- Agent runtime and harness
- Languages
- TypeScript SDK
- Status (October 2026)
- Active
- License
- MIT
- Good fit for
- Persistent sessions, sandboxed tools, and approvals without one model vendor
- Shape
- Self-hosted agent cluster framework
- Languages
- Python
- Status (October 2026)
- Slowing: last release June 2026
- License
- MIT
- Good fit for
- Teams already running it; check activity before adopting
- Shape
- Control plane for agents you already use
- Languages
- Runs Claude Code, Codex, Cursor, OpenClaw, and others
- Status (October 2026)
- Active
- License
- MIT
- Good fit for
- Budgets, approval gates, and an audit trail across several agents
- Shape
- Self-hosted autonomous agent you customize
- Languages
- Runs in Docker; extend with prompts and tools
- Status (October 2026)
- Active
- License
- MIT
- Good fit for
- Tinkerers who want one general agent on their own machine
- Shape
- Domain research framework built on LangGraph
- Languages
- Python
- Status (October 2026)
- Active
- License
- Apache-2.0
- Good fit for
- Multi-agent trading research, not live trading
| Tool | Shape | Languages | Status (October 2026) | License | Good fit for |
|---|---|---|---|---|---|
| LangGraph | Code-first framework (graphs) | Python, JavaScript/TypeScript | Active | MIT | Explicit control over state, branching, and human-in-the-loop checkpoints |
| CrewAI | Code-first framework; visual Crew Studio in the paid AMP platform | Python | Active | MIT | Role-based agent teams that are quick to prototype |
| Microsoft Agent Framework | Code-first framework | Python, .NET; Go in preview | Active; 1.0 since April 2026 | MIT | New Microsoft-stack projects; successor to AutoGen and Semantic Kernel |
| Google ADK | Code-first framework; experimental visual builder (Python) | Python, TypeScript, Go, Java, Kotlin | Active | Apache-2.0 | Teams working in several languages; optimized for Gemini, runs other models |
| Pydantic AI | Code-first framework | Python | Active | MIT | Typed tools and validated, structured outputs |
| Smolagents | Minimal code-first library | Python | Maintained; last release May 2026 | Apache-2.0 | Small agents that act by writing code |
| LlamaIndex | Code-first data and RAG framework | Python (the TypeScript version is deprecated) | Active; the company's focus has shifted to LlamaParse | MIT | Agents grounded in documents and other data |
| Agno | Code-first framework plus AgentOS runtime | Python | Active | Apache-2.0 | Python teams serving agents from their own cloud |
| Mastra | Code-first framework; Agent Builder for stored agents | TypeScript | Active | Apache-2.0 core; enterprise features under a separate license | TypeScript and Node.js applications |
| VoltAgent | Code-first framework plus VoltOps console | TypeScript | Active | MIT | TypeScript multi-agent systems with built-in observability |
| AutoGen | Code-first framework | Python, .NET | Maintenance mode: no new features; Microsoft points new projects to Agent Framework | MIT | Existing AutoGen projects |
| Semantic Kernel | Code-first SDK | C#/.NET, Python, Java | Superseded by Microsoft Agent Framework; v1.x gets critical fixes and security patches | MIT | Existing Semantic Kernel applications |
| Dify | Visual workflow builder | No code needed; self-host or cloud | Active | Source-available (Apache 2.0 with added conditions) | RAG apps and workflows built on a canvas |
| AutoGPT | Visual block builder and chat-built agents | No code needed; self-host or cloud | Active; platform in beta | Platform source-available (PolyForm Shield) | Recurring automations on schedules and webhooks |
| ZooWork | Hosted chat-based agent builder plus managed agent API | No code needed; TypeScript and Python SDKs | Active; Managed Agent API in developer preview | Proprietary | Delivering agents with approval gates to teams and clients |
| Flowise | Visual drag-and-drop builder | No code needed | Discontinued; repository archived August 2026 | Apache-2.0 | Not for new projects |
| Docker Agent | Config-first runtime | YAML or HCL files; Go SDK for embedding | Active | Apache-2.0 | Repeatable agent teams shared through registries like container images |
| TrueForge | Agent runtime and harness | TypeScript SDK | Active | MIT | Persistent sessions, sandboxed tools, and approvals without one model vendor |
| IntentKit | Self-hosted agent cluster framework | Python | Slowing: last release June 2026 | MIT | Teams already running it; check activity before adopting |
| Paperclip | Control plane for agents you already use | Runs Claude Code, Codex, Cursor, OpenClaw, and others | Active | MIT | Budgets, approval gates, and an audit trail across several agents |
| Agent Zero | Self-hosted autonomous agent you customize | Runs in Docker; extend with prompts and tools | Active | MIT | Tinkerers who want one general agent on their own machine |
| TradingAgents | Domain research framework built on LangGraph | Python | Active | Apache-2.0 | Multi-agent trading research, not live trading |
Licenses as stated in each repository; source-available licenses restrict some uses, such as offering the software as a competing service. Editorial comparison, not a ranking.
Common tradeoffs
Code-first control vs visual speed
Code-first frameworks fit version control, tests, and code review, and handle complex logic well. Visual builders let non-developers ship faster but make complex branching and change review harder. The line is blurring: Mastra, CrewAI's AMP platform, and Google ADK add visual tooling, while Dify and ZooWork expose APIs.
Open source vs source-available vs proprietary
Most code-first frameworks here use MIT or Apache 2.0. Dify and the AutoGPT platform are source-available with added conditions, and ZooWork is a proprietary hosted service. Check the license before you build a product or hosted service on top of one.
Substitutes vs complements
Frameworks in the same language are substitutes: LangGraph, CrewAI, and Pydantic AI in Python; Mastra and VoltAgent in TypeScript. A framework, a runtime that serves its agents, and a control plane such as Paperclip can be used together, and gateways, memory, browsers, and sandboxes come from the AI infrastructure layer.
Common questions about agent framework tools
What is the difference between LangGraph and CrewAI?
LangGraph models workflows as explicit graphs with fine-grained state control. CrewAI uses higher-level abstractions — you define agent roles and tasks, and the framework handles coordination. LangGraph gives more control; CrewAI is faster to prototype with. Full comparison →
Is AutoGen still maintained?
AutoGen is in maintenance mode: it receives no new features and is community managed, with contributions limited to fixes and documentation. Microsoft recommends Microsoft Agent Framework, which reached 1.0 in April 2026, for new projects and publishes a migration guide. Semantic Kernel is also succeeded by Agent Framework, with v1.x receiving critical fixes and security patches.
Which framework supports the most programming languages?
Google ADK has official SDKs for Python, TypeScript, Go, Java, and Kotlin. Semantic Kernel supports C#/.NET, Python, and Java, and Microsoft Agent Framework supports Python and .NET, with Go in preview. LangGraph supports Python and JavaScript/TypeScript. CrewAI, Pydantic AI, Agno, and LlamaIndex are Python-only (LlamaIndex's TypeScript version is deprecated), and Mastra and VoltAgent are TypeScript-only.
Are there no-code agent builders in this category?
Yes. Dify builds agents and RAG workflows on a visual canvas, AutoGPT uses a block-based builder and agents described in chat, and ZooWork builds agents from a chat description. Flowise was a popular visual builder but was discontinued in 2026. Several code-first frameworks also add visual tools, such as Mastra's Agent Builder and CrewAI's Crew Studio.
Which framework fits RAG and document-grounded agents?
LlamaIndex is built for retrieval-augmented generation, with data ingestion, indexing, and retrieval at its core; its Python framework is maintained, though the company's main focus has shifted to its LlamaParse platform. Dify includes built-in knowledge pipelines for teams that prefer a visual builder. Other frameworks can do RAG but need more of the retrieval layer built or integrated by you.
Related
- LangGraph vs CrewAI: graphs or role-based crews
The most common choice between two Python frameworks, compared on control, setup, and production use.
- Browse AI infrastructure for agents
The layer underneath: model gateways, browsers, memory, sandboxes, and managed runtimes that framework-built agents use.
- Browse AI coding agents
Ready-made coding agents such as Claude Code and OpenCode, which Paperclip and similar control planes coordinate.
Vendor facts in this guide last checked October 15, 2026 against: LangGraph, LangGraph.js, CrewAI, AutoGen maintenance notice, Semantic Kernel and Agent Framework, Microsoft Agent Framework, Google ADK, LlamaIndex, LlamaIndex.TS notice, Mastra, Dify, Flowise sunset notice, AutoGPT, Docker Agent, Paperclip. Picks and tradeoffs are editorial judgments.