OneWill provides a safety layer that mediates interactions between autonomous AI agents and external systems, offering both sandbox mode with minimal state access and a “free‑range” mode for full production access. The platform records every action, enabling governance features such as provenance tracking, rollback, and reversible execution, so developers retain control over agent behavior while still allowing powerful agency when needed.
Funding
Funding not disclosed
Founders
Product
Problem
Autonomous AI agents often need direct access to user state such as files, email, calendars, or shell commands, but granting unrestricted access creates high risk of data loss, unintended changes, or security breaches. Existing approaches either isolate agents in sandboxed environments with limited usefulness or allow full access without safeguards, leaving users unable to balance functionality and safety.
Solution
OneWill introduces an Agent‑World Boundary that interposes every agent action with a reversible, database‑style transaction layer. Before an action is executed, OneWill records intent and required state changes in a write‑ahead log, ensuring that each operation can be undone or approved in real time. Users retain control through governance policies, provenance tracking, and rollback capabilities, allowing agents to operate on real state while preserving the ability to revert unintended effects. The platform supports both sandboxed and production‑grade modes, giving developers a middle ground that combines useful agency with strong safety guarantees. By leveraging proven crash‑recovery techniques, OneWill provides deterministic, auditable interactions between AI agents and external systems.
Target Audience
Primary customers are developers and enterprises deploying autonomous AI agents that need to interact with sensitive user or system state, such as productivity tool integrations, DevOps automation, and AI‑assisted personal assistants.
Features
- Write‑ahead logging of agent intents to enable atomic UNDO/REDO of state changes
- Real‑time approval workflow for non‑reversible actions, allowing user oversight
- Full provenance metadata for each agent operation, supporting audit and debugging
- Configurable sandbox and production modes to balance access to state versus risk
- Automatic rollback of failed or aborted transactions to preserve original system state
- API layer that mediates agent calls to files, email, calendar, shell, and other external resources