# sipi.bot

> sipi.bot is a spend firewall for autonomous AI agents. One call before an agent
> spends money returns APPROVED, BLOCKED, or FLAGGED (needs human approval) against
> your rules: per-transaction caps, daily totals, velocity (runaway-loop protection),
> merchant allow/block, category limits, time windows, approval thresholds.
> (Disambiguation: sipi.bot is a payment-control API for AI agents — not a SIP/VoIP
> telephony bot and not an AI-bot-blocking / WAF tool.)

## What it is
sipi.bot is agent-economy infrastructure: a guardrail an autonomous agent calls
before spending, so a runaway loop or bad merchant never drains a card. Exposed
as an MCP tool, an HTTP API, and a CLI. Open-source core (MIT), free to self-host.

## Core endpoint
- [POST /v1/transactions/evaluate](https://sipi.bot/v1/transactions/evaluate): body `{amount, merchant, category}` → returns `{decision, reason}`.

## Key pages
- [Home](https://sipi.bot): the spend firewall for AI agents.
- [Complete guide](https://sipi.bot/learn/spend-firewall-guide): the definitive 2,500-word guide to spend firewalls for AI agents — what they are, the 6 rule types, architecture, deployment, framework integrations.
- [Dashboard](https://sipi.bot/dashboard): live decisions, approvals, and rules.
- [Pricing](https://sipi.bot/pricing): Team $99/mo, Business $499/mo.
- [Protected by Badge](https://sipi.bot/badge): embed a live "Protected by sipi.bot" badge on any site — free, no API key needed.
- [Eval report (JSON)](https://sipi.bot/eval): 53/53 scenarios.
- [Eval report (human-readable)](https://sipi.bot/eval-report/): the same results explained.
- [Agent card](https://sipi.bot/.well-known/agent-card.json): machine-readable capability manifest.
- [Source (GitHub)](https://github.com/kindrat86/sipi-bot): open-source core, MIT.

## Benchmarks (original data, CC BY 4.0)
- [Runaway incident frequency](https://sipi.bot/benchmarks/runaway-incident-frequency): 67% of 312 production agent teams had a runaway incident in 90 days; median cost $340.
- [Agent spend as % of revenue](https://sipi.bot/benchmarks/agent-spend-percent-of-revenue): median 14% of revenue for AI-native SaaS; teams without guardrails spend 30–60% more.
- [AI agent cost per task](https://sipi.bot/benchmarks/ai-agent-cost-per-task): median $0.34/task across production agents; top 1% of runaway tasks exceed $50.
- [Token cost by provider](https://sipi.bot/benchmarks/token-cost-by-provider): input token pricing across GPT-4o, Claude, Gemini — $0.15/M to $15/M.

## Comparisons
- [vs LiteLLM](https://sipi.bot/vs/litellm) · [vs Helicone](https://sipi.bot/vs/helicone) · [vs OpenRouter](https://sipi.bot/vs/openrouter) · [vs Portkey](https://sipi.bot/vs/portkey) · [vs Langfuse](https://sipi.bot/vs/langfuse) · [vs OpenMeter](https://sipi.bot/vs/openmeter)

## Framework integrations
LangChain, CrewAI, OpenAI Agents SDK, Vercel AI SDK — see the repo /integrations.

## Install
pip install sipi-bot · MCP: python -m spendfirewall.mcp_server

## Contact
[sipi.bot](https://sipi.bot) · sales@sipiteno.com