Decision API Design for Oracle Wallets: Policies Before Execution
Design a decision API that separates evidence, policy, authorization, and execution, with explicit limits and useful failure states.
Oracle Wallet Lab / Topic
AI Agents
Keep interpretation and authority legible.
Agent-assisted wallet experiences need a defined role for a model, a clear record of its inputs, and limits on any proposed action. These readings examine practical ways to structure that boundary. The AI safety guide focuses on the review process; the decision API guide focuses on a proposal that another component or person can inspect.
Read the two guides together when planning an integration. Start with the task the model should help explain, then write down the permissions it does not need to complete that task. Specify what evidence must accompany a recommendation, when the recommendation expires, and which changes require a new approval. Those questions keep the design discussion grounded in a concrete workflow.
Continue with the AI Oracle Wallet guide for a focused overview and related questions.
Articles in AI Agents
AI Oracle Wallet Safety: From Model Output to Human Approval
A practical architecture for keeping AI interpretation, policy checks, transaction preparation, and human authorization separate.