About sipi.bot

sipi.bot is the spend firewall for autonomous AI agents — the control layer that evaluates every transaction an agent attempts and returns approve, block, or flag before any money moves. Built for the agent economy, open-source (MIT), and free to self-host.

Why we built it

The agent economy handed autonomous software real spending power — API credits, compute, SaaS, payments — usually backed by a human's credit card and no hard limit. Runaway retry loops and unattended purchase chains are failure modes worth testing before an agent receives live payment access. We built sipi.bot because every autonomous agent deserves a guardrail, and every operator deserves to sleep without checking their bank balance.

What sipi.bot does

sipi.bot sits in front of every transaction an autonomous AI agent attempts. One HTTP call — POST /v1/transactions/evaluate — evaluates the proposed spend against your rules and returns APPROVED, BLOCKED, or FLAGGED with a deterministic rules check. Six rule types are enforced: per-transaction caps, daily totals, velocity limits (runaway-loop protection), merchant allow/block lists, category limits, and time-window constraints. Every decision is written to a queryable audit log, and transactions that need human judgment are routed to an approval queue — not auto-approved and not silently blocked.

How it works

You define rules: "Max $500 per transaction," "Max $2,000 per day," "Block any merchant matching *.ru," "Require human approval above $1,000." Your agent calls sipi.bot before it spends money. sipi.bot evaluates every active rule in priority order. The first BLOCK rule that matches stops the transaction instantly — no money moves. FLAGGED transactions enter the human-in-the-loop queue. APPROVED transactions proceed. The entire decision path is logged, timestamped, and auditable.

Surfaces

  • HTTP API — POST /v1/transactions/evaluate with JSON body, returns JSON decision. Works with any agent runtime.
  • MCP Server — Native Model Context Protocol tool. Claude Code, Cursor, and Hermes call evaluate_transaction directly.
  • CLI — pip install sipi-bot && sipi-guard for shell scripts, CI pipelines, and cron jobs.
  • Agent Card — /.well-known/agent-card.json for A2A (Agent-to-Agent) discovery.

Pricing

Flat-rate, no per-call fees. Team: $99/month. Business: $499/month. Both include unlimited transaction evaluations. The open-source core is MIT-licensed and free to self-host forever — same rule engine, same latency, no limits.

Who's behind it

sipi.bot was built by Maryan, a solo founder and AI infrastructure engineer based in Kifisia, Greece. Maryan has been building in the agent-economy stack since 2024 — spend controls, payment protocols (x402, AP2), MCP tooling, and compliance infrastructure. Previously built sanctions screening tools for AI agent payments (sanctionsai.dev) and churn analytics (churnlens.site).

Open source

The sipi.bot core is open-source under the MIT license. The rule engine, MCP server, CLI, and evaluation logic are all public at github.com/kindrat86/sipi-bot. The hosted version layers billing, dashboard, and approval-queue persistence on top of the same engine.

← Back to sipi.bot