Nunchuk has released two open-source repositories aimed at reshaping how AI agents interact with Bitcoin wallets, introducing a model that limits agent control while preserving human oversight.
The release includes Nunchuk CLI, a command-line interface for managing Bitcoin wallets, and a companion “Agent Skills” repository designed to help AI systems operate the CLI across common workflows. Both tools are licensed under MIT and target developers building automated financial systems on Bitcoin, Nunchuk said.
The core premise challenges a growing trend in AI wallet design. Rather than granting agents full control over funds with basic safeguards, Nunchuk proposes a shared custody model where agents operate within strict policy limits. Human users retain final authority over transactions that exceed predefined thresholds.
Under this structure, wallets are configured as group wallets with multiple keys. A user key, an agent key, and a policy co-signer work together to authorize transactions. The agent can initiate actions such as creating wallets, inviting participants, and constructing transactions, but spending authority remains constrained by rules set at the policy level.
AI-driven bitcoin use
These policies define limits such as daily spending caps, approval requirements, and signing delays. Transactions that fall within allowed parameters can proceed without intervention, while larger or sensitive actions require explicit user approval.
Nunchuk separates custody from automation. The wallet structure governs ownership and control of funds, while policy layers define what an agent can execute. This distinction ensures that funding a wallet does not grant broader authority to the agent managing it.
The CLI supports a range of functions including key generation, wallet creation, transaction workflows, and policy configuration. It also allows users to export wallet descriptors and backups using standard formats, which supports portability and recovery outside the Nunchuk ecosystem.
The Agent Skills repository serves as an interface layer for AI systems. It provides predefined commands and prompts that guide agents through tasks such as setting up wallets, managing participants, and executing transactions. This reduces the need for custom integrations and lowers the barrier for developers experimenting with Bitcoin-based automation.
Nunchuk positions the dual-repository approach as a response to two distinct challenges: execution and usability. The CLI acts as the execution layer tied to the Nunchuk API, while the skills layer focuses on how AI systems interact with that infrastructure.
The release reflects a broader effort to define safe design patterns for AI in financial contexts. By enforcing bounded authority, Nunchuk aims to enable practical automation without introducing full custodial risk.
Potential use cases include shared human-agent wallets, automated bill payment systems, treasury management tools, and multi-agent coordination. While these applications remain in early stages, the underlying model provides a framework for controlled experimentation.
As AI systems gain access to financial tools, the question of control becomes central. Nunchuk’s approach suggests that the path forward may depend less on restricting capability and more on structuring authority.
Editorial Disclaimer: We leverage AI as part of our editorial workflow, including to support research, image generation, and quality assurance processes. All content is directed, reviewed, and approved by our editorial team, who are accountable for accuracy and integrity. AI-generated images use only tools trained on properly license material. In Bitcoin, as in media: Don’t trust. Verify.