Flow AI Information

Flow builds AgentAuth, the control plane for what AI agents do. Your agents keep their own model, planning, and memory. When one acts, whether it calls a tool, moves money, or sends mail, AgentAuth verifies who is asking, judges the action against the intent you approved, rehearses the write before it commits, and puts every step on a tamper evident record. Revocable instantly. Most agent security stops at a valid token. A fully authorized agent can still be prompt injected, misled by a poisoned tool response, or simply wrong. AgentAuth governs the action itself: deterministic policy decides what may happen, and AI judgment can narrow or refuse an action but never expand authority. A wrong or manipulated model degrades to more caution, never less. Agents hold no credentials. Each approved action runs on a fresh, narrowly scoped, short lived token, on open standards (AAuth, OAuth 2.1, MCP). Flow also ships AI agents for regulated banking operations and an AI native ERP generated from your own spreadsheets and processes, all on the same control plane. Built by AI researchers and software architects from Google and Apple. Now onboarding early access teams.