Agent Spend Guardrails for Platform Engineers
You run the platform agents run on. If they spend, that's your pipeline to control — and prompts aren't controls.
The platform problem
Every team ships agents; every agent can spend. You need a shared layer, not per-team hope.
The guardrail has to be deterministic — no model call that can be prompt-injected or argued with.
It must fit your stack: HTTP, MCP, or CLI.
What platform teams get
One API call (or MCP tool) that returns APPROVED, BLOCKED, or FLAGGED in ~5 ms.
Rules as code: caps, allowlists, velocity limits, time-of-day rules.
MIT-licensed core — self-host inside your VPC if you need to.
How to deploy
Put sipi.bot in front of your payment path. Agent → sipi.bot → rail. Wire the audit log into your observability stack.
What you get
| Platform concern | sipi.bot answer |
|---|---|
| Latency | ~5 ms decision, no model call |
| Lock-in | MIT core, self-host anywhere |
| Fit | HTTP API, MCP tool, CLI |
| Audit | Queryable rule-level log |
FAQ
Is it deployable in a private network?
Yes — the core is MIT-licensed and self-hosts; the decision engine runs fully offline with your rules.
Does it work with our gateway?
Yes — sipi.bot composes with LiteLLM, Cloudflare AI Gateway, and similar proxies: firewall first, then route.
What's the operational footprint?
Small — a single service with a SQLite store; unlimited evaluations on hosted plans.
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