{% extends "base.html" %} {% block styles %}{{ super() }}{% endblock %} {% block content %}
Open Source · MIT Python 3.11+ LangGraph

Autonomous AI agents
that see and act.

Python SDK for agents that control the desktop, browse the web, and call any tool — with built-in safety, memory, and human approval.

Get started → See guides
Install pip install gantrygraph
agent.py 
from gantrygraph import GantryEngine, gantry_tool
from gantrygraph.perception import DesktopScreen
from gantrygraph.actions import MouseKeyboardTools
from langchain_anthropic import ChatAnthropic

@gantry_tool
async def notify(channel: str, msg: str) -> str:
    """Post a message to a Slack channel."""
    return await slack.post(channel, msg)

agent = GantryEngine(
    llm=ChatAnthropic(model="claude-sonnet-4-6"),
    perception=DesktopScreen(),
    tools=[MouseKeyboardTools(), notify],
    max_steps=30,
)
agent.run("File the quarterly report and ping #finance")
✓ Opened Excel  ·  ✓ Exported PDF  ·  ✓ Posted to #finance
What it does

Everything your agent needs

Composable primitives. No framework lock-in.

See and act on any UI

Screenshot the desktop or a browser, decide what to click, type, or navigate — no APIs needed.

Desktop guide →

Call any tool or service

Wrap a function with @gantry_tool or connect GitHub, Postgres, Notion via MCP — two lines.

Build tools →

Safety and control

Workspace sandboxing, human approval gates, budget caps. Know exactly what runs before it runs.

Add guardrails →
{% endblock %}