langchain-valta

Hard spending limits for LangChain agents. Enforces budget policies before any paid operation fires.

Install

pip install langchain-valta

Quickstart

from langchain_valta import ValtaBudgetTool
from langchain_openai import ChatOpenAI
from langchain.agents import initialize_agent, AgentType

guard = ValtaBudgetTool(
    api_key="vk_live_...",   # from valta.co/dashboard/api-keys
    agent_id="my-agent",
)

agent = initialize_agent(
    tools=[guard, ...your_other_tools],
    llm=ChatOpenAI(model="gpt-4o"),
    agent=AgentType.OPENAI_FUNCTIONS,
    agent_kwargs={
        "system_message": (
            "Before any paid operation, call check_spend with the estimated cost. "
            "Only proceed if the result is 'approved'."
        )
    },
)

How it works

The agent calls check_spend before any paid operation. Valta evaluates the request against the user's policy and returns approved or denied: <reason> before the charge fires.

Policy management

Set limits at valta.co/dashboard/policies. Changes take effect immediately — no redeploy needed.

Links

• valta.co
• Documentation
• Get an API key