gavrun 0.2.5

Python SDK for Gavrun — govern AI agents before risky actions execute

gavrun

Python SDK for Gavrun — govern AI agents before risky actions execute.

Gavrun intercepts agent tool calls, evaluates them against your policies, and requires human approval for high-risk actions before they run.

Install

pip install gavrun

Quick start

import gavrun

gavrun.configure(
    api_key="your-sdk-key",       # from agentmanager.gavrun.ai/api-keys
    agent_name="my-agent",
    agent_manager_url="https://api.gavrun.ai",
)

Tool calls are governed automatically for LangChain, LangGraph, CrewAI, and OpenAI Agents SDK via auto-patch. For manual guarding:

client = gavrun.get_client()

result = client.guard(
    action="delete_user",
    execute=lambda: delete_user(user_id),
    tools=["delete_user"],
    prompt=f"Delete user {user_id}",
)

if result.denied:
    raise PermissionError("Blocked by policy")
if result.requires_human_review:
    # poll for approval
    ...

Frameworks

Auto-patch wires governance into every tool call:

import gavrun
gavrun.configure(api_key="...", agent_manager_url="https://api.gavrun.ai")
# LangChain, LangGraph, CrewAI, OpenAI Agents SDK tools are now governed

Links

• Dashboard
• gavrun.ai