Nunchuk, a Canadian bitcoin wallet provider focused on multisignature solutions, has unveiled two new open-source projects aimed at creating safer ways for AI agents to interface with Bitcoin wallets. The company, which has built its reputation on secure, policy-driven wallet frameworks, now seeks to redefine how artificial intelligence can be used in automated financial workflows while keeping users firmly in control of their assets.
New approach to AI wallet access
The newly launched projects include the Nunchuk CLI tool and an “Agent Skills” repository. Both are licensed under MIT and designed for developers who want to build or test automated use cases involving Bitcoin without exposing funds to excessive risk. Nunchuk’s CLI allows command-line management of wallets, while the Agent Skills repository serves as a toolkit for AI agents working through standard wallet tasks.
This release reflects a shift from traditional AI wallet design that typically gives agents broad custodial access to user funds. Instead, Nunchuk proposes a system where AI agents operate under strict, predefined rules, keeping human users as the ultimate authority for sensitive or high-value transactions.
Wallets following this design are configured as group wallets, utilizing signatures from multiple entities—a user, an agent, and a policy co-signer. This structure requires joint approval for actions that surpass established limits, such as spending above daily thresholds or changing key wallet settings.
Developers are able to leverage the CLI for tasks like generating keys, creating and managing wallets, defining transaction workflows, and establishing custom policy controls. Export options ensure compatibility and recovery beyond just the Nunchuk ecosystem.
AI agent skills and human oversight
The Agent Skills repository offers predefined commands and templates that AI agents can use, making it easier for developers to automate wallet management without the need for building complex integrations from scratch. This separation between wallet custody and AI automation creates a balance where routine tasks can be automated but authority over funds remains with the user.
All agent actions are limited by custom policy parameters including spending caps, approval steps, and delayed signatures for large transactions. Any actions falling outside policy require human authorization, significantly reducing security risks compared to fully autonomous agents.
According to Nunchuk, this combination of layered control is designed to address two major challenges in AI-crypto integration: safe execution and broad usability. The CLI handles low-level operations, while the skills layer enables AI interaction following established safety guidelines.
Nunchuk sees real-world applications for these tools in areas such as treasury management, automated payment scheduling, and collaborative finance, where balancing automation and oversight is critical. The company highlights that these use cases are still early in their development but expects growing interest as AI becomes more deeply integrated with financial platforms.
By defining a model where AI capabilities are explicitly separated from wallet control, Nunchuk aims to support innovation in Bitcoin-based automation without increasing custodial risk. The approach could set a precedent for how other companies design agent-driven financial tools moving forward.
