Shotoku
Shotoku is open-source, local-first authorization infrastructure for AI agents. It evaluates agent actions before they happen, enforcing budgets, policy rules, human approvals and a tamper-evident audit trail on the user's machine without holding funds or keys.
At a Glance
- Developers building autonomous AI agents
- Teams operating MCP-compatible agents such as Claude Desktop, Claude Code and Cursor
- Customer-support and commerce teams using agentic refunds or payments
- Organizations needing governance, compliance and auditability for agents
- +1 more
AI Tools by Shotoku
(1)Shotoku
Open Source AI Agent Spend Control
Discussions
No discussions yet
Be the first to start a discussion about Shotoku
Latest News
Shotoku featured in Show HN: Local-first action governance and budget caps for AI agents
Shotoku published current npm packages: shotoku-cli 0.2.0, shotoku-mcp 0.2.0 and @shotoku/core 0.3.0
Repository work introduced the clickable hosted dashboard demo and PayPal evaluation workflow
How to set a hard spending limit on an AI agent
Products & Services
Open-source TypeScript authorization and policy engine exposing authorize() and x402 authorization, with deterministic policy evaluation, budgets, fail-closed behavior and a hash-chained decision ledger.
Command-line interface for initializing policies, checking authorization, inspecting status, approving or denying requests, running demos and operating the local terminal/dashboard workflows.
MCP server that can operate in advisory mode or wrap upstream MCP servers so tool calls are policy-checked before forwarding; it supports approval queues and tool pricing.
Integration that wraps PayPal's MCP server, prices refunds and dispute claims, applies limits and routes over-limit actions to human approval without taking custody of PayPal funds.
Market Position
Shotoku positions itself as a local-first, deterministic policy engine and audit layer rather than a hosted AI gateway, virtual card, custodial agent wallet or billing-alert product. Its differentiators are enforcement before the action, human approvals outside the agent, persistent tamper-evident local records and no custody of money or keys. The company's own comparison identifies AgentPay MCP, x402-cfo and PolicyLayer as adjacent x402 spend-control alternatives; Shotoku emphasizes its local policy/ledger approach and built-in human approval.
Leadership
Founders
Julius Peschard
French design engineer and product designer; his public profile describes end-to-end product design and engineering, and he is identified as Shotoku co-founder with responsibility for product and go-to-market.
Issa Prunier
French computer-science student and AI-security/red-teaming builder. His public profile describes him as an AI-security practitioner and co-founder building Shotoku.
Executive Team
Julius Peschard
Co-founder
Design engineer and product designer responsible for product and go-to-market, according to his public profile.
Issa Prunier
Co-founder
Computer-science student and AI-security/red-teaming builder who publicly describes himself as building Shotoku.
Founding Story
Shotoku started in June 2026 as a policy engine for x402 payments. The founders then expanded it during the PayPal AI Hackathon after October 1, 2026, adding PayPal MCP integration, pricing of money-moving tools, human approval flows and the hosted dashboard demo. The initial vision was to give agents a deterministic authorization layer that runs before an action or payment, rather than relying on prompts, alerts or a custodial wallet.
Business Model
Revenue Model
The open-source authorization core is free and runs on the customer's machine. Shotoku's planned monetization is paid shared-control and governance functionality for teams and enterprises, while policies and the ledger remain on the customer's infrastructure.
Pricing Tiers
authorize() core and policy engine, hash-chained ledger, CLI and terminal UI, MCP server, x402 authorization, signed snapshots and verification.
Shared control for teams running multiple agents across projects; planned organization-level policy hierarchy, Slack approval routing, multi-approver rules and a central ledger dashboard.
Planned governance, enforcement and compliance features including signed receipts, RBAC, compliance export packs, SSO/SAML/SCIM and support SLA.
Target Markets
- Developers building autonomous AI agents
- Teams operating MCP-compatible agents such as Claude Desktop, Claude Code and Cursor
- Customer-support and commerce teams using agentic refunds or payments
- Organizations needing governance, compliance and auditability for agents
- Users and developers of x402 and other machine-payment rails
- Controlling AI-agent purchases and paid API calls
- Gating MCP tools and destructive database or infrastructure actions
- AI customer-support agents issuing refunds with thresholds and daily caps
- x402 stablecoin/pay-per-request payments
- Checking agent email recipients before sending
- Auditing and reviewing autonomous agent actions
- PayPal is the featured integration and sandbox use case; the project wraps PayPal's MCP server in its support-agent refund demonstration.