An open-source safety layer that checks every AI agent action before it runs
Emerged finds new products worth a look, daily — what’s new today
Shotoku
Spending rules and human sign-off for agents, checked on your own machine
Is this yours? Claim itBefore you let an agent near a payment method, this gives you per-vendor limits, an approval step and a log without handing anyone your funds.
Shotoku checks an agent's spending request against rules you write before any payment happens. One YAML file sets per-transaction and rolling 24-hour limits per vendor; requests over the line are denied or held until a person approves, from Telegram, a terminal UI or a small dashboard. It never holds money, stores credentials or executes payments. Decisions land in a JSONL log on disk, and approvals produce signed, short-lived tokens a payment proxy or CI can verify. MIT, in beta, from two developers in France who plan an enterprise version.
More like this
A local ledger of every tool call your coding agents make
Approval gates and an audit ledger for coding agents left running unattended
A credential vault and approval gate between AI assistants and your apps
One key and one interface for 2,000+ agent tools across 72 providers
Inventory every AI agent on your machine and what it can reach