AI agent frameworks
Frameworks provide the scaffolding that lets AI agents plan, call tools, keep state and work together – from LangChain, LangGraph and DeepAgents to CrewAI, AutoGen and Hermes Agent.
An agent framework provides the building blocks of an AI agent: connections to language models, tool calling, memory for intermediate state, control of multi-step workflows and often collaboration between several agents.
Frameworks differ mainly in their level of abstraction. Some ship an opinionated way of working out of the box (a so-called agent harness); others are low-level runtimes in which you model every step yourself.
Role in an AI-first business model
In an AI-first business model, the framework is the technical foundation of the operating model: it determines how well mandate, approvals, escalations and traceability can be represented. If you design processes from the outcome for agents and people, you need a framework that supports human-in-the-loop, durable state and recovery after failures.
For AI-native business models, where agents carry the core of value creation, the choice is even more fundamental. It shapes operating costs, model independence and how quickly the system can evolve.
What to look for when choosing
- Controllability: can workflows, permissions and stop points be defined explicitly?
- Human-in-the-loop: can people review, edit or reject actions before they run?
- State and recovery: do long-running workflows survive failures and interruptions?
- Model independence: can you switch the language model without rebuilding the application?
- Observability: are decisions and tool calls traceably logged?
62 tools in this category
LangChain
Open-source framework for AI agents and LLM applications with a unified interface to models, tools, vector stores and many integrations – for Python and JavaScript/TypeScript.
LangGraph
Low-level framework by LangChain for orchestrating long-running, stateful agents as a graph – with durable execution, human-in-the-loop interrupts and short- and long-term memory.
DeepAgents
Open-source agent harness by LangChain built on LangGraph: planning, sub-agents, a filesystem, context management, skills and approvals for tool calls come built in.
Hermes Agent (Nous Research)
Self-improving open-source agent by Nous Research, the team behind the Hermes language models: it creates skills from experience, maintains memory, works via messengers and CLI and runs with any model provider.
OpenClaw
Open-source framework for persistent personal AI agents. It connects language models to messaging channels such as Signal, Telegram or WhatsApp and adds scheduling, memory, MCP integration, skill plugins, sub-agents and browser automation.
Transformers Agents
Agent interface from the Hugging Face ecosystem that adds natural-language control on top of the Transformers library. Useful for combining models and tools from that ecosystem through an agent in Python.
LlamaIndex
Framework that connects language models to external data sources. A typical use is RAG over company data: ingesting and indexing documents and giving agents or applications targeted access to them.
Botpress
Platform providing building blocks for chatbots and conversational agents. Suited to designing dialogue flows, enriching them with language models and deploying them across different channels.
Haystack
Python framework by deepset for composing pipelines from language models, search and data sources. Commonly used for RAG, question answering and agents that work with your own data.
Semantic Kernel
Microsoft SDK for integrating language models into your own applications, including in C#. Existing code can be exposed as plugins and used by agents via tool calling.
AGiXT
Platform for automation with AI agents that follow instructions and organise multi-step tasks. Intended for scenarios in which agents work through more complex processes on their own.
LLM Agents
Compact Python project for building agents that are controlled by a language model and call tools. Well suited to understanding the basic mechanics of an agent.
e2b
Open-source platform for building and deploying agents that handle development tasks. It provides isolated cloud environments in which agents can run generated code safely.
Dust
Platform for designing large language model applications and deploying them to production. Suited to assistants that work with company knowledge and connected tools.
MetaGPT
Multi-agent framework that models the roles of a software team. From a one-line requirement, the agents produce a requirements document, design, tasks and a code repository.
Lagent
Lightweight framework from the InternLM project for building agents based on language models. Suited to developers looking for a lean foundation for tool calling and agent logic.
AutoGen
Microsoft framework for applications in which several agents communicate and solve tasks together. Applicable to multi-agent orchestration with tools and human involvement.
AG2
Open-source framework from the creators of AutoGen for building AI agents. Its focus is on cooperation between multiple agents that divide up and solve tasks together.
AgentVerse
Flexible framework for setting up custom multi-agent environments for language models. Useful for team-based task solving by agents and for simulating how they interact.
Maestro
Framework in which Claude Opus acts as the orchestrating model, breaking tasks into subtasks and handing them to sub-agents. The results are then combined into an overall result.
AgentScope
Framework intended to simplify building multi-agent applications based on language models. Suited to defining several agents with clear roles and managing how they interact.
CrewAI
Python framework for orchestrating autonomous agents that work together as a team ("crew") with defined roles. Typical for divided workflows such as research, analysis and writing.
Swarm
Educational framework from OpenAI exploring lightweight multi-agent orchestration. It demonstrates how agents can pass tasks to one another through simple handoffs.
agency-swarm
Agent framework built on top of the OpenAI Assistants API. It lets you combine several agents with their own roles and tools into a collaborating "agency".
Upsonic
Agent framework focused on reliable execution that supports the Model Context Protocol (MCP). Suited to connecting agents to external tools and data sources via MCP.
Mastra
Opinionated TypeScript framework for building AI applications and features quickly. It covers agents, workflows and tool integration, among other things.
Vectara-agentic
Python framework for creating AI assistants and agents on top of the Vectara platform. Suited to agents that combine RAG over your own data with tool calling.
AgentDock
Open-source foundation for building, managing and deploying AI agents and workflows in production. It aims to reduce the effort of handling many separate API integrations.
Swarms Framework
Framework for orchestrating many agents, aimed at enterprise applications. It offers different patterns by which agents work on tasks together, in parallel or in sequence.
Strands Agents SDK
SDK that takes a model-driven approach to building agents: the language model plans and chooses tools itself. A working agent can be set up in just a few lines of code.
VoltAgent
Open-source TypeScript framework for building AI agents. Built-in LLM observability makes it possible to trace what agents are doing in operation.
Agentic Context Engine
Framework for agents that learn from feedback on their own executions and curate their own context. An integration with LangChain is available.
Astron
A platform from iFlytek for agentic workflows, aimed at enterprise use. It helps teams model multi-step processes with AI agents and run them under a commercially friendly licence.
Ailoy
A framework for AI agents meant to run in many environments. With support for local models and WebAssembly, it suits agents that work without a cloud connection or directly in the browser.
PraisonAI
A multi-agent framework with agent self-reflection, workflows, memory and MCP integration. It offers SDKs for Python and JavaScript and works with a wide range of language models.
AgentField
Open-source infrastructure for AI backends that combines orchestration and identity management for agents. It is aimed at teams that want to run agents as long-lived services in production.
Cortex Memory
A memory solution for agents that extracts information, retrieves it via vector search and maintains the store automatically. It is accessible through a REST API, MCP and a CLI, plus a dashboard.
Pipecat
An open-source Python framework for voice and multimodal conversational AI. It is useful for building voice agents that combine speech recognition, a language model and speech output in real time.
LoongFlow
A framework from the Baidu ecosystem for developing self-evolving agents. It spans atomic building blocks, the development framework itself and ready-made agents for core scenarios.
Agentset
An open-source RAG platform with agentic reasoning, hybrid search and multimodal support. It helps agents answer questions grounded in a company’s own data.
Pilot Protocol
A network stack written in Go that gives AI agents virtual addresses and encrypted connections, including NAT traversal. It lets agents on different machines communicate directly with mutual trust.
hcom
A tool that lets coding agents such as Claude Code, Gemini CLI, Codex or OpenCode message, watch and spawn each other across terminals. It supports multi-agent collaboration on the command line.
nanobot
A lightweight Python framework for a personal AI assistant with a compact codebase. It supports MCP, several chat channels and an extensible skills system.
Hive
An open-source framework for goal-driven autonomous agents: outcomes are described in natural language and turned into agent graphs that improve through evolution loops. It includes MCP integration and a large set of tools.
ConnectOnion
A simple Python framework in which plain functions serve as tools. Lifecycle hooks, a plugin system and trust-based networking between agents make it usable for production agents.
SwarmClaw
A self-hosted runtime for multi-agent orchestration with schedules, delegation, memory and skills. It connects coding agents and various LLM providers and runs as a desktop app or CLI.
smolagents
An open-source Python library from Hugging Face for building and running agents with very little code. It suits quick prototypes and agents that express their actions as code.
Open Multi-Agent
TypeScript-native multi-agent orchestration that plans tasks as a dependency graph at runtime and executes them in your own environment. It provides approval gates, tracing, evaluation and checkpoints with resume.
Aeon
A framework for autonomous agents that run unattended via GitHub Actions. It offers self-healing skills, output quality scoring, persistent memory and reactive triggers.
Cordum
A governance control plane that runs outside the agents and enforces policies before an action is dispatched. Approval gates and signed audit trails support oversight of autonomous agents in production.
trpc-agent-go
A Go framework for production agent systems with graph workflows, tools and memory. It supports protocols such as A2A, AG-UI and MCP and includes evaluation and observability.
Agentlas-OS
A local-first, model-agnostic environment for building, borrowing and orchestrating specialist agents and agent teams. Packages are owner-scoped, and permissions are enforced by the host environment.
Lobu
An open-source context and control plane for AI agents: company tools are connected once and their context is shared, respecting permissions, across different assistants and custom agents. It includes MCP, scoped identities, approvals and audit.
OpenProgram
An agent framework in which agents create and refine their own workflows. It offers multi-agent orchestration, persistent sessions, sandboxed execution, and desktop and web interfaces.
SandBase Harness
A local-first managed runtime for agents with persistent sessions and sandboxed execution. MCP tools, audit, replay of past runs and a human console support controlled operation.
Agent Swarm
A self-hosted multi-agent system in which a lead agent delegates tasks to specialised workers. Shared memory, tools, schedules and review gates keep the work traceable.
agent-sdk-go
A Go framework for building stateful AI agents. It suits teams that want to integrate agents directly into existing Go services.
OpenHuman
A local-first personal AI agent for the desktop with agents, workflows and an MCP client. Language models can run locally via Ollama.
Tenuo
Authorization for agent tools that checks every single call down to the argument level. Authority remains traceable across delegation chains between agents.
Pydantic Deep Agents
A Python framework built on Pydantic AI with planning, sub-agents, file and shell tools, and MCP. Agents can optionally run in a Docker sandbox, and it also ships as a terminal coding agent.
CUGA
A generalist agent harness from IBM Research built on LangGraph that executes tasks over APIs and the web. It provides policy and human-in-the-loop controls, supervision of remote A2A agents and agent skills.
Omnara
An open-source platform for running durable agents on any model and machine, from sandboxes to your own laptop. It includes MCP, skills, subagents and human approvals, and it can be self-hosted.
From tool to outcome
In AI-first coaching you clarify which outcomes AI agents should be accountable for in your go-to-market and within which framework. REVAGO then designs, builds and runs your system. You pay for results – not for activity.
Key concepts
Other categories
A neutral overview, not a recommendation of individual providers. REVAGO is not affiliated with the listed projects. Descriptions are based on the projects’ public information; features and licences may change. Selection based on awesome-agents (kyrolabs), As of 19 October 2026.