A factual definition from the agent wallet directory
A crypto wallet whose keys and transactions are controlled by code, enabling software and AI agents to hold funds and sign automatically under predefined rules.
A programmable wallet is a cryptocurrency wallet whose key management and transaction logic are exposed through application programming interfaces (APIs) or software development kits (SDKs). This architecture allows software—including AI agents, backend services, and automated scripts—to hold digital assets, construct transactions, and produce valid cryptographic signatures without requiring a human to manually approve each operation through a traditional wallet interface. The "programmable" aspect refers to the ability to embed conditional logic, spending rules, and automated workflows directly into how the wallet operates.
Traditional wallets—whether hardware devices, browser extensions, or mobile apps—center on human interaction. A user opens an interface, reviews a transaction, and confirms it with a password, biometric scan, or physical button press. Programmable wallets invert this model: the signing capability is abstracted into software-accessible services that can be invoked by code.
The underlying mechanisms vary by implementation. In MPC (multi-party computation) systems, a private key is mathematically split into multiple shares distributed across different servers or parties. No single machine ever holds the complete key, yet they can jointly produce signatures. In smart account models, the wallet itself is a smart contract on-chain, with programmable access rules, spending limits, and recovery procedures encoded in its logic. Custodial programmable wallets rely on a service provider to manage keys on behalf of the user or application, trading self-custody for operational simplicity.
Regardless of the specific architecture, the common thread is that transaction authorization flows from code-defined rules rather than from real-time human discretion. An agent might check an oracle price, verify a condition, and execute a swap—all without pausing for manual confirmation. This enables use cases like automated treasury management, machine-to-machine payments, and AI agent economies where speed and continuity matter.
The agentwallet.md directory lists multiple programmable wallet implementations with different custody models and specializations:
A programmable wallet is a cryptocurrency wallet whose key management and transaction signing are exposed through APIs or SDKs, allowing software—including AI agents—to hold funds and execute transactions according to code-defined rules rather than direct human action.
Regular crypto wallets require human interaction for key management and transaction approval, typically through browser extensions or mobile apps. Programmable wallets expose these functions programmatically, enabling automated, conditional, or agent-initiated transactions while maintaining cryptographic security.
Programmable wallets employ several custody models: MPC (multi-party computation) splits keys across multiple parties without ever reconstructing them; smart accounts use programmable contract logic for access control; and custodial models rely on a service provider to manage keys. Each model involves different trust assumptions and security trade-offs.
Programmable wallets enable autonomous agent payments, recurring transactions, conditional execution based on external data, micropayments between machines, and on-chain automation where software must interact with blockchain networks without waiting for human approval at each step.
Security depends on the implementation. MPC-based wallets reduce single points of failure by distributing key shares. Smart accounts can include recovery mechanisms and spending limits. However, any programmable system introduces attack surfaces related to code bugs, API compromise, and oracle manipulation that do not exist with simple manual wallets.