guard()

guard() wraps a tool so Kelhe scores the action before it runs. There is one adapter per framework; the call shape is the same, except LangGraph and AutoGen take a shared AgentContext.

from kelhe.adapters.openai_agents import guard   # OpenAI Agents
from kelhe.adapters.langgraph import guard, AgentContext   # LangGraph
from kelhe.adapters.crewai import guard           # CrewAI
from kelhe.adapters.autogen import guard, AgentContext     # AutoGen

Framework	Call	Notes
OpenAI Agents	guard(kelhe, func)	Returns a FunctionTool.
CrewAI	guard(kelhe, tool)	Function or BaseTool; scoped by tool name.
LangGraph	guard(kelhe, func, context=ctx)	Create one AgentContext, then ctx.bind(agent).
AutoGen	guard(kelhe, tool, context=ctx)	Create one AgentContext, then ctx.bind(agent).

All adapters accept name= and description= to override what the model sees. The tool's behavior is never changed — Kelhe only runs a check immediately before it executes. See the per-framework guides under Getting started.