Spend Control for Semantic Kernel
Semantic Kernel is Microsoft's agent framework for .NET, Python, and Java — powering enterprise automations with plugins and planners. Enterprise agents spend on paid plugins, and nobody budgets the plugins.
Why Semantic Kernel agents overspend
Plugins can call paid endpoints; a planner chaining plugins multiplies calls.
Enterprise deployments run many agents on shared credentials — one runaway loop hits the shared bill.
There's no built-in budget concept in the kernel's plugin pipeline.
How it works
Add sipi.bot as a plugin (or call the HTTP API from a custom plugin). Before any spend, the kernel asks sipi.bot: APPROVED, BLOCKED, or FLAGGED — deterministic, ~5 ms.
Rules that fit enterprise workloads
Shared daily ceiling across the agent fleet.
Merchant allowlist for approved paid plugins and vendors.
Approval threshold so large purchases route to a human.
Plugin call
// Semantic Kernel plugin (C#)
var result = await client.PostAsJsonAsync(
"https://sipi.bot/v1/transactions/evaluate",
new { amount = 800, merchant = "paid-plugin.com", category = "api" });
var decision = (await result.Content.ReadFromJsonAsync<Decision>()).decision;
// APPROVED | BLOCKED | FLAGGED
Gate any plugin that spends.
FAQ
Does sipi.bot support .NET?
Yes — sipi.bot is a plain HTTP API and MCP tool, so .NET, Python, and Java Semantic Kernel apps all call it directly.
Can it cap a whole enterprise fleet?
Yes — per-agent rules plus a shared daily ceiling govern the fleet; the audit log attributes every decision.
Does it add latency to planners?
About 5 ms per check, no model in the path.
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