figuard-langchain 1.2.0

FiGuard pre-flight spend authorization for LangChain agents

figuard-langchain

Pre-flight spend authorization for LangChain agents. One line to add budget enforcement and velocity controls to any AgentExecutor.

pip install figuard-langchain

Quick start

from figuard_langchain import auto_guard_langchain

# Monetary budget — enforces dollar spend on tools with an "amount" parameter
executor = auto_guard_langchain(executor, budget=500)

# Velocity control — catches runaway loops even when tools have no dollar amount
executor = auto_guard_langchain(executor, budget=500, velocity_max_per_minute=10)

result = executor.invoke({"input": "Book a flight to NYC"})

auto_guard_langchain creates a FiGuard budget, wires a callback handler to the executor, and returns the same executor ready to run. Uses the shared public sandbox by default — no account required.

What gets enforced

Scenario	How to enforce
Tools with dollar amounts (price, cost, amount fields)	budget=500 — denies when total spend exceeds $500
Research agents, code agents, any tool without a financial parameter	velocity_max_per_minute=10 — denies the 11th call in 60 seconds
Both	Pass both params — monetary ceiling + loop detection

OTEL / Langfuse integration

FiGuard emits authorization spans to any OpenTelemetry backend automatically when one is configured. With Langfuse:

from langfuse import Langfuse
from figuard_langchain import auto_guard_langchain

lf = Langfuse(public_key="...", host="https://us.cloud.langfuse.com")
executor = auto_guard_langchain(executor, budget=500, velocity_max_per_minute=10)

Every authorize() call appears as a GUARDRAIL span in the Langfuse trace. Denied calls appear as ERROR spans — visible in the execution graph without any extra configuration.

Full control

For advanced fleet patterns, delegation tokens, or per-category allocations, use the main figuard package directly:

pip install figuard[langchain]

from figuard import FiGuardClient
from figuard.integrations.langchain import FiGuardCallbackHandler

client = FiGuardClient()
budget = client.create_budget(
    user_id="alice",
    total_limit=500.00,
    currency="USD",
    velocity_max_per_minute=10,
    allocations=[
        {"category": "flights", "limit": 300.00, "mode": "CATEGORY_CONSTRAINED"},
        {"category": "hotels",  "limit": 200.00, "mode": "CATEGORY_CONSTRAINED"},
    ],
)
handler = FiGuardCallbackHandler(
    client=client,
    session_token=budget.primary_token.session_token,
)
executor.callbacks = [handler]

Self-hosting

git clone https://github.com/figuard/figuard-core
cd figuard-core
docker compose up -d

Set FIGUARD_BASE_URL=http://localhost:8080 and FIGUARD_API_KEY=<your-key> before running your agent.

Links

• GitHub
• Documentation
• Public sandbox
• PyPI (figuard)