Spend Control for AWS Bedrock Agents
Bedrock Agents build on AWS's LLM platform — and AWS bills you for the inference. But the agent's purchases beyond AWS (data vendors, external APIs, compute) are spend Bedrock never sees.
Why Bedrock Agents overspend
Inference volume: agent loops and tool calls multiply token spend on your AWS bill.
Agents can call external paid tools (data, compute, APIs) outside Bedrock's pricing entirely.
Cost explorer shows the bill after the fact — nothing gates a transaction in real time.
How it works
Call sipi.bot from your agent's action group before any spend. The decision — APPROVED, BLOCKED, or FLAGGED — is deterministic and logged.
Rules that fit Bedrock workloads
Per-agent daily ceiling for inference.
Merchant allowlist for approved external vendors.
Category rule: inference, data, compute budgets stay separate.
Action group call
import requests
def guard(amount, merchant, category):
r = requests.post(
"https://sipi.bot/v1/transactions/evaluate",
json={"amount": amount, "merchant": merchant, "category": category},
headers={"Authorization": "Bearer KEY"},
)
return r.json()["decision"] # APPROVED | BLOCKED | FLAGGED
Wire it into the action group that spends.
FAQ
Does this replace AWS Budgets?
No. AWS Budgets alerts on your AWS bill; sipi.bot decides before each agent transaction — including spend outside AWS. They compose.
Can it cap Bedrock inference specifically?
Yes — category rules give inference its own ceiling; the agent can't exceed it regardless of what it's doing.
Is the decision instant?
About 5 ms, no model call.
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