Every request follows the same path: prepare context, reason about the task, then loop through tool calls until the work is done. This is the core processing logic behind every Arlo run — in the TUI, the one-shot CLI, or the library.
A turn starts from a user prompt, a scheduled task, an event or webhook, a notification, or an MCP request. Whatever the source, it enters the same pipeline — nothing gets a private code path.
Each step streams as an event, which is what the TUI renders live and what an embedding application can subscribe to. See how it's built →