Spend Control for Letta
Letta (formerly MemGPT) agents keep memory across sessions — which means they keep calling tools across sessions. Long-lived agents need long-lived budgets.
Why Letta agents overspend
Persistent agents call tools repeatedly across days.
Memory-heavy workflows re-run paid lookups.
Autonomous background agents spend without review.
How it works
Attach the guard as a tool. Before any spend — lookup, purchase, subscription — the agent gets a deterministic decision.
Rules that fit Letta workloads
Per-agent daily ceiling (persistent agents spend daily).
Merchant allowlist for paid tools.
Approval threshold for new vendor spend.
Guard call
from sipi_guard import sipi_guard
# Inside a tool that spends:
decision = sipi_guard(amount=25, merchant="enrichment-api.com", category="data")
# APPROVED | BLOCKED | FLAGGEDLong-lived agents, enforced budgets.
FAQ
Do persistent agents need different rules?
Yes — they spend daily by design, so daily ceilings fit better than one-off caps.
Is the decision model-free?
Yes — deterministic rules, ~5 ms.
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