Once

Building Once — AI agent execution safety for consequential tool calls, retries, ambiguous outcomes, and MCP workflows. Open source at onceexec.com.