Shotoku logo

Shotoku

Local spending limits, human approvals and an audit trail for AI agents.

Shotoku product preview

The product

What Shotoku is building

Shotoku is a command-line tool that puts spending controls in front of AI agents. Its official site describes budgets and rules written in a single YAML policy file, covering actions such as purchases and paid API calls, with per-request and daily limits. Each request is checked before the payment happens and receives one of three verdicts: approved, pending approval or denied. Larger spends can wait for a person to approve them. The site says decisions are recorded on the user's own machine in a tamper-evident log, that there is no cloud service involved and that Shotoku does not take custody of funds. It installs through npm, and the site links to documentation, a changelog and source code.

From the maker

Give an agent a budget instead of a card

As agents begin to call paid APIs and make purchases, the question of who approved a charge becomes a finance problem as much as an engineering one. Shotoku answers it with a policy file and a local decision log checked before money moves. It was presented on Hacker News on 14 November 2026, and the site invites design partners, which signals an early product. It is worth a look for teams testing agents that can spend.

Community

Launch discussion

Say who you are, what clicked, and what still feels unclear.

Be the first person to leave thoughtful feedback.