verifiedx 0.1.1

VerifiedX Python SDK

VerifiedX Python SDK

Minimal Python SDK for framework-native VerifiedX capture and boundary preflight.

Install

pip install verifiedx

Set VERIFIEDX_API_KEY in the app environment. VERIFIEDX_BASE_URL is optional and only needed for self-hosting or non-default environments.

Quickstart

from verifiedx import init_verifiedx

vx = init_verifiedx()


class AgentHarness:
    def call_model(self, messages):
        return {"messages": messages}

    def send_email(self, to, body):
        return {"to": to, "body": body, "status": "sent"}


harness = AgentHarness()
vx.install_runtime(
    harness,
    llm={"call_model": {"model_name": "gpt-5.4-mini"}},
    tools={"send_email": {}},
)

init_verifiedx() stays stable. The default integration path is now a single install_runtime(...) bind that auto-instruments supported Python runtime surfaces, captures conversation windows, and routes boundary decisions through the V4 raw-capture lane without builder-authored semantic callbacks.

LangGraph: from verifiedx.langgraph import compile

LangGraph native adapter, few LOC:

from verifiedx import init_verifiedx, install_langgraph
from langgraph.graph import StateGraph

vx = init_verifiedx()
install_langgraph(verifiedx=vx)

builder = StateGraph(MyState)
# add nodes / edges as usual
graph = builder.compile(checkpointer=my_checkpointer, store=my_store)

That path keeps LangGraph-native capture and trigger naming for Command(update=...), graph.update_state, store.get/search/put/delete, checkpointer.put/put_writes, LangGraph tool execution through BaseTool.invoke/ainvoke, AIMessage.tool_calls, and ToolMessage.tool_call_id.

LangChain: from verifiedx.langchain import create_agent

OpenAI direct: from verifiedx import install_openai_direct

OpenAI Agents: from verifiedx import install_openai_agents

Anthropic direct: from verifiedx import install_anthropic_direct

Claude Agent SDK: from verifiedx import install_claude_agent

Doctor: verifiedx doctor