The open-source agent framework
A modular, async-first multi-agent framework for Python. Type-safe tools, swarm orchestration, and 14 composable packages — from prototype to production.
Everything you need, nothing you don't
A single Agent class that scales from quick scripts to production multi-agent systems.
Single Agent Class
One composable Agent — no inheritance hierarchies. Configure with models, tools, instructions, and handoffs.
Type-Safe Tools
The @tool decorator auto-generates JSON schemas from type hints. Full validation, zero boilerplate.
Multi-Agent Swarms
Workflow, Handoff, and Team orchestration modes for complex multi-agent coordination.
Modular Packages
14 focused packages — install only what you need. From core to MCP, memory, sandbox, and more.
Async-First
run(), run.sync(), run.stream() — every execution mode you need, with first-class async support.
Production-Ready
MCP servers, structured output, tracing, evaluation, human-in-the-loop — built for real workloads.
Up and running in minutes
Define a tool, create an agent, and run it. That's it.
# hello.py — your first Orbiter agent
from orbiter import Agent, tool, run
@tool
def get_weather(city: str) -> str:
"""Get the current weather for a city."""
return f"It's sunny in {city}!"
agent = Agent(
name="weather-bot",
instructions="You are a helpful weather assistant.",
tools=[get_weather],
)
result = run(agent, "What's the weather in Tokyo?")
print(result.output)
# => It's sunny in Tokyo!14 packages, one ecosystem
A layered architecture where each package has a single responsibility. Install only what you need.
Built for teams of agents
Three orchestration modes for every multi-agent pattern you need.
Workflow
Sequential pipeline execution with a Flow DSL. Chain agents together with deterministic ordering and data passing.
Handoff
Agent-driven delegation. Agents decide which specialist to route to based on context, with automatic tool generation.
Team
Lead-worker pattern with auto-generated tools. A lead agent coordinates specialists and synthesizes results.
Start building with Orbiter
Open-source, MIT licensed, and ready for production.
pip install git+https://github.com/Midsphere-AI/orbiter-ai.git