Spend Control for Replicate
Replicate hosts open models billed per prediction. The guard caps the line for agents running models at volume.
Why Replicate agents spend
Per-prediction billing at agentic volume.
Image and model experiments multiply calls.
How it works
Call the guard before predictions: amount, merchant, category → APPROVED, BLOCKED, or FLAGGED.
Rules that fit
Per-agent daily ceiling on model calls.
Category rule: image vs language models.
Velocity limit on retry loops.
Guard call
import requests
r = requests.post("https://sipi.bot/v1/transactions/evaluate",
json={"amount": 0.8, "merchant": "replicate", "category": "models"},
headers={"Authorization": "Bearer KEY"})
decision = r.json()["decision"]Predictions are a bill; the gate makes it a budget.
FAQ
Does this slow model calls?
No — ~5 ms per check.
Can I cap per project?
Yes — per-agent rules per project.
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