The Open-Source Framework for Agentic AI
OpenAgents is the open-source Python framework for building and deploying multi-agent AI systems. Connect agents across networks with native MCP & A2A protocols for discovery, communication, and real-time collaboration.
$ pip install openagents
$ openagents init my-agent-network
$ openagents deploy
✓ Network deployed — 3 agents connected
Everything You Need to Build an Agent Platform
OpenAgents provides the full stack for multi-agent systems — from agent identity and discovery to network orchestration and deployment. Built open source so you own every layer.
Agent Networks
Create networks where hundreds or thousands of AI agents collaborate in real time. Each network defines its own rules, roles, and communication patterns for multi-agent coordination.
AgentID Registry
Every agent gets a verifiable AgentID — a decentralized identity for discovery and trust. Other agents and humans can look up capabilities, verify provenance, and establish secure connections.
MCP & A2A Protocols
First-class support for Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication. Agents can share context, delegate tasks, and coordinate using open, standardized protocols.
Modular Extensions
Extend the platform with community-built Mods. Add new LLM integrations, custom tools, data connectors, and agent behaviors without touching the core framework.
Visual Studio
Design agent workflows visually in OpenAgents Studio. Drag-and-drop agents into networks, configure connections, monitor real-time activity, and debug agent interactions — no code required.
Deploy Anywhere
Run the agent platform on your own infrastructure or use OpenAgents Cloud. Docker-native deployment, Kubernetes-ready, with full control over your data and agent behavior.
Open Architecture for Agent Interoperability
The OpenAgents platform is built on open standards so your agents can communicate with any other agent — not just ones built on our framework. No vendor lock-in, no proprietary protocols.
Agent Layer
Individual agents with their own LLM backends, tools, and memory. Each agent registers an AgentID for discovery and verification.
Protocol Layer
MCP for sharing context and tools, A2A for direct agent-to-agent communication. Standard HTTP and WebSocket transports for universal compatibility.
Network Layer
Agent Networks define the collaboration topology — orchestrated, peer-to-peer, or hybrid. Built-in load balancing and fault tolerance for production workloads.
# OpenAgents Platform Architecture
# ─────────────────────────────────
network:
name: "my-agent-network"
type: orchestrated
agents:
- id: "researcher"
model: gpt-4o
tools: [web_search, arxiv]
protocols: [mcp, a2a]
- id: "analyst"
model: claude-sonnet
tools: [python, charts]
protocols: [mcp, a2a]
- id: "writer"
model: gemini-pro
tools: [markdown, publish]
protocols: [mcp]
connections:
- from: researcher
to: analyst
channel: findings
- from: analyst
to: writer
channel: insights
deploy:
target: cloud # or: self-hosted
replicas: 3
autoscale: trueWhat Developers Build on the Platform
From research automation to production software systems, the OpenAgents platform powers multi-agent applications across every industry.
Multi-Agent Research Teams
Deploy networks of specialized agents that search, analyze, and synthesize information across thousands of sources in parallel. One agent finds papers, another extracts data, a third generates insights.
Agent Network
Autonomous Dev Pipelines
Build agent networks that handle entire development workflows — from reading issues to writing code, running tests, and opening pull requests. Agents coordinate through A2A to ship faster.
Agent Network
Intelligent Data Pipelines
Orchestrate agents that ingest, clean, transform, and analyze data at scale. Each agent specializes in a step of the pipeline and passes structured outputs to the next agent via MCP.
Agent Network
Start Building in Minutes
Go from zero to a running agent network in three steps. The platform handles identity, discovery, and communication so you can focus on what your agents actually do.
Install the CLI
Install the OpenAgents CLI with pip. Works on macOS, Linux, and Windows (WSL).
pip install openagentsCreate Your First Agent
Define an agent with a model, tools, and protocols. The platform handles identity registration and discovery automatically.
from openagents import Agent
agent = Agent(
name="my-agent",
model="gpt-4o",
tools=["web_search"],
)Join or Create a Network
Connect your agent to an existing network or create a new one. Agents discover each other through the AgentID registry.
from openagents import Network
network = Network("research-team")
network.add(agent)
network.run()Frequently Asked Questions
Common questions about the OpenAgents platform, agent networks, and getting started.
An open agent platform is a software framework that lets you build, deploy, and connect AI agents using open-source code and open protocols. Unlike closed platforms, an open agent platform gives you full control over your agents, data, and infrastructure. OpenAgents is built on open standards like MCP (Model Context Protocol) and A2A (Agent-to-Agent), so your agents can interoperate with any compatible system.
Most agent frameworks focus on single-agent or small-team scenarios. OpenAgents is designed specifically for agent networks — large-scale systems where hundreds or thousands of agents collaborate. Key differentiators include: built-in AgentID for decentralized identity, native MCP and A2A protocol support, a visual Studio for designing workflows, and community-driven Mods for extending functionality.
OpenAgents is model-agnostic. You can use any LLM backend — OpenAI (GPT-4o, o1), Anthropic (Claude), Google (Gemini), open-source models (Llama, Mistral, Qwen), or your own fine-tuned models. Different agents in the same network can use different models, letting you optimize for cost, speed, or capability per task.
Yes. OpenAgents is fully open source under the MIT license. You can run the entire platform on your own infrastructure using Docker or Kubernetes. This gives you complete control over data privacy, security, and compliance. We also offer OpenAgents Cloud for teams that prefer a managed deployment.
Every agent registered on the platform gets an AgentID — a verifiable identity that includes the agent's capabilities, supported protocols, and connection endpoints. Other agents can query the AgentID registry to discover agents by capability, then establish direct communication channels using MCP for context sharing or A2A for task delegation.
OpenAgents is actively used in production by teams building research automation, data processing pipelines, and software development workflows. The platform includes built-in monitoring, error handling, and autoscaling. Join our Discord community to see how other teams are deploying agent networks at scale.
Connect with AI Agent Network Builders
Join thousands of developers, researchers, and AI enthusiasts building the future of agent networks. Get help, share ideas, and collaborate on cutting-edge AI Collaboration.
