Framework Adapters

Gavrun ships adapters for LangChain, LangGraph, CrewAI, and OpenAI Agents SDK. When auto_patch=True (the default), configure() monkey-patches your framework automatically.

Auto-patch (recommended)

import gavrun

gavrun.configure(
    api_key="gavrun_live_...",
    agent_manager_url="https://api.gavrun.ai",
    auto_patch=True,   # default
)

# Every LangChain / LangGraph / CrewAI tool call is now governed

Manual patch

gavrun.configure(auto_patch=False)

from gavrun.adapters import patch_frameworks
patch_frameworks()

LangChain / LangGraph

Auto-patch wraps BaseTool.invoke, BaseTool.run, and BaseChatModel.invoke. No code changes needed.

from langchain_core.tools import tool

@tool
def delete_order(order_id: str) -> str:
    """Delete an order."""
    return db.delete(order_id)

# delete_order is now governed — calls preflight before executing

Use LangChainPreflightRunnable for explicit control:

from gavrun.adapters import LangChainPreflightRunnable

guarded = LangChainPreflightRunnable(
    client=gavrun.get_client(),
    runnable=my_tool,
    payload_factory=lambda input: client.guard_payload(
        action="delete_order",
        tools=["delete_order"],
        prompt=f"Delete order {input}",
    ),
)
result = guarded.invoke(order_id)

CrewAI

Auto-patch wraps BaseTool.run. No code changes needed.

For manual control use CrewAIPreflightTool:

from gavrun.adapters import CrewAIPreflightTool

guarded_tool = CrewAIPreflightTool(
    client=gavrun.get_client(),
    tool=my_crewai_tool,
    payload_factory=lambda args: client.guard_payload(
        action="web_search",
        tools=["web_search"],
        prompt=f"Search: {args}",
    ),
)

OpenAI Agents SDK

Use OpenAIPreflightHooks as lifecycle hooks:

from gavrun.adapters import OpenAIPreflightHooks
from agents import Agent, Runner

hooks = OpenAIPreflightHooks(
    client=gavrun.get_client(),
    payload_factory=lambda tool_name: client.guard_payload(
        action=tool_name,
        tools=[tool_name],
        prompt=f"Tool call: {tool_name}",
    ),
)

agent = Agent(name="my-agent", tools=[...])
result = await Runner.run(agent, input="...", hooks=hooks)