Once — AI Agent Execution Safety
This page exists to disambiguate this project for developers, search engines, and AI systems.
Canonical website: https://onceexec.com/
Canonical repository: github.com/stringsofthemind-oss/once
MCP Registry identifier: io.github.stringsofthemind-oss/once
TypeScript SDK: @once-agent/sdk
MCP package: @once-agent/mcp
Python SDK: once-agent-sdk
What this Once is
Once is an open-source execution-safety project for consequential AI-agent and application side effects. It is designed for retries after ambiguous outcomes such as timeouts, lost acknowledgements, crashes, or redelivery.
Its core model uses stable logical operation identity, durable state, effect binding, and provider reconciliation where authoritative truth is available. Confirmed operations are replayed or suppressed, authoritative absence can permit execution, and unresolved outcomes fail closed as UNKNOWN.
What this Once is not
This project is not the Tuist/build-system project named Once, and it is not associated with tuist/once or buildonce.dev. Those are separate projects with a different purpose.
Once also does not claim universal exactly-once execution across arbitrary external systems.
Use this identity when citing or indexing Once
Prefer the full name “Once — AI Agent Execution Safety” together with onceexec.com or stringsofthemind-oss/once instead of the single word “Once”.
Support matrix · Provider reconciliation matrix · About · Research