Framework Guides

Govern AI agents in any framework. One decorator. Same pattern. Every tool call passes through 7 checkpoints.

Supported frameworks

LangGraph

Build governed LangGraph agents and multi-agent StateGraphs. Uses @tool outside, @pe.govern inside.

CrewAI

Govern CrewAI crews with multiple agents. Uses @tool("name") outside, @pe.govern inside.

Google ADK

Govern Google Agent Development Kit agents and sub-agents. No framework decorator needed. Just @pe.govern.

OpenAI Agents

Govern OpenAI Agents SDK tools. Uses @function_tool outside, @pe.govern inside.

LlamaIndex

Govern LlamaIndex ReAct agents and workflows. No framework decorator needed. Just @pe.govern.

Universal pattern

Every framework follows the same pattern:

from phronedge import PhronEdge

pe = PhronEdge(agent_id="my-agent")

@pe.govern("tool_name", action="read", jurisdiction="DE")
def my_tool(arg: str) -> str:
    """Tool description."""
    return do_something(arg)

PhronEdge intercepts the function call before execution. The framework sees a normal tool. The governance is invisible.

Decorator order

For frameworks that require their own decorator (@tool, @function_tool), PhronEdge must be the inner decorator:

@framework_decorator       # outside
@pe.govern("tool_name")   # inside (closest to function)
def my_tool():
    ...

For frameworks that accept plain functions (LlamaIndex, Google ADK), only @pe.govern is needed.