Short, practical walkthroughs for the day-to-day work of running a spend firewall — configuring allowlists, setting per-agent limits, wiring up alerts, debugging a block, and integrating with the agent runtimes you already use.
A merchant allowlist restricts which merchants your AI agent can transact with. This tutorial covers allowlist strategy, common patterns, and how to test i
When sipi.bot blocks a transaction, the audit trail tells you exactly which policy rule fired. This tutorial walks through reading the audit log and a
Claude Code can rack up $40 in 20 minutes when it loops. This tutorial walks through inserting sipi.bot as a middleware that caps per-loop spend and s
sipi.bot can alert you on threshold breaches, velocity spikes, and policy changes. This tutorial covers alert routing (Slack, email, webhook) and thre
Walk through configuring per-agent spend limits in sipi.bot: define a per-transaction ceiling, a daily cap, and a velocity cap. Verify with a test tra
LangChain agents call tools; tools spend money. Adding the guard is a few lines — here's the exact walkthrough.
CrewAI runs crews of role-based agents in parallel — which compounds spend. The guard tool gives the whole crew a budget.
n8n workflows run on schedules — including at 2 AM. A firewall node before any paid step keeps the automation honest.
Category rules give each spend type its own budget. Here's the setup in three steps.
The classic runaway happens at 2 AM. A time-of-day rule makes unattended overnight spend require a human.
Approval queues keep agents fast while material purchases wait for a human. Setup in three steps.
The core is MIT-licensed and self-hosts anywhere Docker runs. Here's the container walkthrough.
The AI SDK makes streaming AI features easy — and easy features get agents that spend. The guard slots into the same flow.
The daily ceiling is the single most useful budget number. Setup in three steps.