Spend Control for Google ADK
Google's Agent Development Kit builds multi-agent systems on Gemini. Multi-agent means multi-spend — subagents fan out, tools get called, and the bill follows.
Why ADK agents overspend
Subagent fan-out multiplies inference and tool calls.
Tools can trigger paid services (Vertex APIs, data, payments) without a budget check.
Retry and reflection loops compound spend.
How it works
Attach sipi.bot as a tool in your ADK agent. Before spending, the agent calls the guard and gets a deterministic decision.
Rules that fit ADK workloads
Per-agent daily ceiling for the whole fleet.
Category rule: inference vs data vs payments.
Velocity limit to stop retry loops.
Guard tool
from sipi_guard import sipi_guard
from google.adk.agents import Agent
agent = Agent(name="spending_agent", tools=[sipi_guard])
# decision = sipi_guard(amount=300, merchant="vertex-api", category="inference")One tool, deterministic policy.
FAQ
Does this work with Gemini pricing controls?
Gemini/Vertex controls cap that provider's inference. sipi.bot governs the agent's purchases across every merchant — run both.
Is there a Python SDK?
Yes — HTTP API, CLI, and MCP tool, plus thin client wrappers for common frameworks.
What about subagents?
Shared ceilings and velocity limits apply fleet-wide, so a fan-out can't compound one mistake.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check