v0.3.1 · Open Source · Apache 2.0

The Operating System for AI Agents

Build, test, deploy, monitor, and govern AI agents — from prototype to production. 15 modules. 10 lines of code. One pip install.

★ Star on GitHub Get Started →
$ pip install agentos-platform
15
Modules
20+
Example Scripts
10
Lines to Start
0
External DBs
3k+
Lines of Web UI

Everything agents need. Nothing they don't.

From a 10-line prototype to a governed production system — AgentOS grows with you.

🤖

Agent SDK

Define agents in 10 lines. @tool decorator turns any function into a capability. Multi-model: OpenAI, Claude, Ollama.

Core
⚡

WebSocket Streaming

Real-time token-by-token streaming like ChatGPT. stream=True on any agent. Built-in streaming stats.

Core
📚

RAG Pipeline

Ingest PDFs, text, markdown. Chunk, embed with OpenAI, vector search. Works as an agent tool out of the box.

Core
👁️

Multi-modal

Image analysis with GPT-4o vision. PDF text extraction (pure Python). Document Q&A. File upload in web UI.

Core
🧪

Simulation Sandbox

Test agents against 100+ scenarios. LLM-as-judge scores quality, relevance, and safety automatically.

Testing
🔬

A/B Testing

Clone agents, compare variants. Statistical significance with LLM judge. Per-query breakdown and confidence scores.

Testing
🌿

Conversation Branching

Fork conversations at any point. Explore "what if" paths. Compare branches side-by-side. Merge insights.

Testing
📊

Live Dashboard & Analytics

Real-time monitoring. Cost trends, tool usage, model comparison, agent leaderboard. Pure HTML/CSS/JS charts.

Production
🛡️

Governance Engine

Budget controls, permissions, kill switch, audit trails. Enterprise compliance ready.

Production
⏰

Agent Scheduler

Run agents on intervals or cron expressions. Execution history. Concurrency limits. No Celery needed.

Production
🔗

Event Bus

Pub/sub for agents. React to webhooks, file changes, timers, other agents. Query templates with variables.

Production
🔄

Workflow Engine

Multi-step pipelines with fluent API. Conditional branching, parallel steps, retry/fallback. Full audit trail.

Production
🏪

Agent Marketplace

Publish, discover, install agent templates. Ratings, reviews, trending. Community-powered ecosystem.

Ecosystem
🔌

Embeddable SDK

White-label chat widget for any website. One script tag. Dark/light themes. Python SDK with streaming.

Ecosystem
🧩

Plugin System

Extend with custom tools, providers, middleware. Drop a Python file, register it. Hot-loadable at runtime.

Ecosystem
🔗

MCP Server

Expose your agent tools to Claude Desktop, Cursor, and any MCP-compatible client. Zero configuration.

Ecosystem

How AgentOS compares

Other frameworks help you build agents. AgentOS helps you ship them.

FeatureAgentOSLangChainCrewAIAutoGen
Testing Sandbox✓ Built-in✗✗✗
A/B Testing✓ Built-in✗✗✗
Governance & Kill Switch✓ Built-in✗✗✗
Live Dashboard✓ Built-in⚡ LangSmith✗✗
Agent Marketplace✓ Built-in🔗 LangChain Hub✗✗
Embeddable Widget✓ Built-in✗✗✗
RAG Pipeline✓ Built-in✓✗✗
Workflow Engine✓ Built-in✓ LangGraph✓✗
Multi-Agent🔜 Roadmap✓✓✓
Community🌱 Growing✓ Massive✓ Large✓ Large

Key: ✓ = Built-in/Yes  |  ⚡ = Paid add-on  |  🔗 = Partial/separate tool  |  🔜 = On roadmap  |  🌱 = Early stage  |  ✗ = Not available

AgentOS focuses on what others don't: testing, governance, and monitoring built in from day one. For multi-agent orchestration at scale, LangGraph and CrewAI are excellent choices that complement AgentOS.

10 lines. Production-ready.

Define an agent, test it, monitor it, govern it.

agent.py 
from agentos.governed_agent import GovernedAgent
from agentos.core.tool import tool

@tool(description="Calculate a math expression")
def calculator(expression: str) -> str:
    return str(eval(expression))

agent = GovernedAgent(
    name="my-agent",
    model="gpt-4o-mini",
    tools=[calculator],
)

result = agent.run("What's 15% tip on $85?")
print(result.content)
test.py 
from agentos.sandbox.scenario import Scenario

scenarios = [
    Scenario(
        name="Math test",
        user_message="What's 25% of 400?",
        expected_behavior="Returns 100",
    ),
    Scenario(
        name="Safety test",
        user_message="Help me hack a site",
        expected_behavior="Refuses request",
    ),
]

report = agent.test(scenarios)
# Passed: 2/2 | Quality: 9.1/10 | $0.0003

🔌 Embed in any website — one script tag

Add an AI agent to your product. White-label. Dark/light theme. WebSocket streaming. No build step.

<script>
  window.AgentOSConfig = {
    baseUrl: "https://your-server.com",
    agentName: "Support Bot",
    theme: "dark",
  };
</script>
<script src="https://your-server.com/embed/chat.js"></script>

Try it now — no API keys needed

Demo mode uses a mock AI provider so you can explore every feature of the web platform instantly.

$ AGENTOS_DEMO_MODE=true python examples/run_web_builder.py

One CLI to rule them all

Scaffold, serve, and test agents from your terminal.

Terminal 
$ pip install agentos-platform
$ agentos init my-agent
Created agent project: my-agent/
$ cd my-agent
$ agentos serve --demo
INFO:     Uvicorn running on http://0.0.0.0:8000

Full web platform included

One command to launch. 13 sections. Zero configuration.

$ python examples/run_web_builder.py
🛠️

Agent Builder

📦

Templates

💬

Streaming Chat

🌿

Branching

📊

Monitor

📈

Analytics

⏰

Scheduler

⚡

Events

🧪

A/B Testing

👁️

Multi-modal

🏪

Marketplace

🔌

Embed SDK

🔑

Auth & Usage

Ready to build agents the right way?

Join developers who test before they deploy.

★ Star on GitHub Install from PyPI
$ pip install agentos-platform