Policy-driven spending controls for AI agents, with local-first YAML limits, duplicate detection, and approvals. MIT licensed and free to self-host.
ActOnce Spend Gateway is an open source, local-first decision engine that controls how AI agents spend money. As autonomous agents such as OpenClaw, Hermes, Claude Code, and Cursor start making real purchases, instructions like be careful with spending stop being a safety mechanism, because an agent can ignore, misread, or forget them between sessions. The gateway sits between the agent and the payment step: it evaluates every transaction against a YAML policy file and returns a single decision of approved, rejected, or requires approval. It ships as an MCP server and command line tool, installs with one npm command, and never stores card numbers, Stripe credentials, or payment tokens.
YAML policy engine: Define maximum purchase amounts, merchant allowlists, category blocks, business hours, and approval thresholds in a readable config file that the agent cannot override at runtime.
Duplicate detection: A SQLite-backed dedupe window stops the same purchase from being approved twice in a configurable time period, which addresses the common double domain registration and repeated subscription problem.
Human-in-the-loop approvals: Purchases above your threshold are queued instead of executed, and you receive a Telegram alert where you can approve or decline before the agent continues.
MCP-native integration: Built as a Model Context Protocol server, so it connects to any MCP-compatible agent through a short JSON config rather than custom glue code.
Zero credential exposure: The gateway only emits decisions. Your payment method, Stripe Link credentials, and tokens stay outside the tool entirely, and the open source edition runs fully on your own machine with no external API keys.
Developers running autonomous agents who need hard limits on what an agent can buy, plus dry-run mode to test policies before enabling real spending.
Solo builders and agencies managing agents across several client projects, where per-project caps, approval queues, and clear audit logs replace guesswork.
Security, finance, and operations leads who must answer why a purchase happened and need a timestamped record of the policy matched, the approval, and the reason.
No. It is a decision engine only. It returns yes, no, or wait, and a payment tool such as Stripe Link handles the actual transaction. The gateway never touches your card or payment credentials.
The open source version is entirely local. Policy files, logs, and the deduplication database live on your machine with no external services. A hosted dashboard with approvals, multi-agent management, and webhooks is available on the paid Pro, Team, and Business plans.
Any MCP-compatible agent. It is tested with OpenClaw and Hermes and works with Claude Code, Cursor, and other frameworks that support the Model Context Protocol.
AppUFO
Translate Xcode .xcstrings into 36+ languages with AI using AppUFO. Preserve placeholders, control glossary, and export back to Xcode. Ship faster globally.
VibeBot
Create AI-powered Discord bots without coding. Describe features in plain English, get instant deployment. Build custom moderation, music, tickets & more.
LLM Search Console
LLM Search Console: track citations and brand mentions in ChatGPT, Claude, Gemini, Perplexity & more. Measure AI visibility, share of voice, and crawl-to-mentio
Content24
Content24 is an AI marketing platform that lets you create, plan, and publish content across all formats from one workspace.
Perry
Maker
Share