How Fork AI Works
Fork accepts explicit requests through ask and can participate in configured AI channels. Channel routing decides whether a message should receive a response; the provider generates text and may request available tools.
A request’s path
Section titled “A request’s path”Message → response routing → permitted context → provider → authorized tools → delivered response.
The model does not become a server administrator. A tool request passes separate catalog, current-message intent, member permission and guild-policy checks. Some high-impact workflows require their own confirmation or exact-plan approval.
Context is bounded
Section titled “Context is bounded”Channel history, server FAQ and personal memory are distinct sources. They can be unavailable or limited, and the model may still answer incorrectly. A stored fact does not imply that semantic indexing is enabled. See Context and memory.
What counts as a result
Section titled “What counts as a result”“I will do that” is not an execution receipt. For moderation, scheduling or server changes, inspect the command confirmation, stored record or step result. An uncertain external result should be reviewed before replay.
Provider availability and usage depend on the running deployment. These docs do not promise a fixed model or unlimited access. See Provider availability and Usage and limits.