Add Decrypt as your preferred source to see more of our stories on Google.
In brief Nunchuk released two open-source tools designed to let AI agents interact with Bitcoin wallets under strict limits.
The system uses shared wallets and approval policies so agents cannot spend funds beyond defined rules.
The tools aim to support automated financial tasks while keeping humans in control of private keys.
Bitcoin wallet company Nunchuk has released open-source tools that allow AI agents to interact with Bitcoin wallets without giving them full control over funds.
Nunchuk said in a blog post on Wednesday that the tools aim to support AI agents that can carry out financial tasks such as sending payments or managing wallets, while ensuring agents do not control the keys that secure users’ Bitcoin.
“Current approaches either give the agent full authority over a standalone wallet, or use delegated signing where the agent acts on the user’s behalf,” Nunchuk founder and CEO Hugo Nguyen told Decrypt. “The problem with both is the same: once the agent is set up, there’s no meaningful check on what it can do with your money. If it’s compromised, misconfigured, or just makes a bad call, nothing stops it.”
Launched in 2020, Nunchuk is an open-source mobile Bitcoin wallet that uses multisignature security to support self-custody, inheritance planning, letting users store Bitcoin with multiple keys instead of relying on a single private key.
The company said its new software release is published under an MIT open-source license and includes two repositories: Nunchuk CLI and Agent Skills for Nunchuk CLI. Nunchuk CLI is a command-line tool that lets AI agents interact with Bitcoin in shared wallets where users keep control of their private keys. If a transaction goes over a set spending limit, the user must approve it.
“The agent can operate within bounded authority, but above the limit, the human still has to sign,” Nguyen said. “Just as importantly, funding the wallet and authorizing the agent are separate decisions: the wallet can receive funds without automatically increasing what the agent is allowed to spend. That’s the design problem we’re solving.”
Agent Skills provides an interface that allows AI models to use the CLI across tasks such as wallet setup, wallet creation, participant invitations, policy configuration, and transactions.
The system separates receiving Bitcoin from spending authority. Depositing funds does not give an AI agent permission to spend them.
“That separation matters,” Nunchuk said. “Funding a wallet and authorizing an agent should not be the same decision, and in this model, they aren’t.”
Spending is controlled through policies such as limits, approval steps, or delays. The tools use shared Bitcoin accounts known as multisignature wallets, which require more than one key to approve a transaction.
Nguyen said the tools are intended for developers building systems that combine AI automation with human oversight, arguing that giving agents direct access to crypto funds creates unnecessary risk.
“They’re the ones who feel the gap most acutely,” he said. “They want to give their agents real financial capabilities, but handing them an unconstrained wallet isn’t something they’re comfortable with.”