Automation

Usage

Clipboard OCR → Translate

Overview

Tuff's screenshot OCR flow as an automation you can edit and run: an AI re-read when native OCR isn't confident enough, then a translation back to the clipboard.

• The first time it scrolls into view it runs once (about 3.3 seconds), and the edges the data travels along move; reduced motion shows the finished run with no timers.
• Click a node to edit it for the next run; set Native OCR's minimum confidence to 94% or more (the sample reads 93%) and the run detours through the AI re-read.
• With Simulate rate limit (429) on the Translate node, translation fails first and succeeds on a retry 600ms later.
• Stop marks the step in progress as cancelled and adds a cancelled by user log line.
• Nodes drag and snap to the dot grid; a drag is kept as an offset, so it survives full screen.
• Wide (≥ 960px) shows the run history, four-step progress, and the log pane; in the column and on narrow screens, Node and This run become tabs.

Customization

• For your own flow, nodes and edges are plain arrays and flowNodes() only computes coordinates; edges always run from a node's bottom to the next node's top, so offset branch nodes sideways.
• For a real runner, replace the timed steps in buildPlan() with backend events (SSE or WebSocket); each event calls setNode() once and appends a log line.
• TxFlowchart has no per-edge styling API: the template borrows dashed for "in flight" and keeps the color and motion inside prefers-reduced-motion: no-preference.
• TxFlatSelect and the fine-tune card's type menu don't teleport; keep them clear of overflow: hidden ancestors and away from the container's bottom edge.

Related components