Use MojoFlow with AI Workbench
Dictate straight into the AI Workbench chat, start dictation from the workbench's microphone button, and share the workbench's language model.
MojoFlow and AI Workbench work together on the same computer. Both are on by default: AI Workbench opens its side of the link whenever MojoFlow is installed, and MojoFlow finds it on its own.
Dictate into the workbench chat
With the workbench's chat box focused in VS Code (or Cursor, Windsurf or VSCodium), hold your key and speak: the text goes straight into the chat box instead of being typed key by key. If the chat box is not focused, or the workbench does not answer within a moment, MojoFlow types the text as usual, so nothing is lost.
You can also click the microphone button in the workbench chat. It starts a dictation, and clicking it again stops it; the text lands in the workbench chat whichever app is in front.
Under Dictation → Mojo Up AI Workbench, MojoFlow lists the workbenches it has found and whether each is connected. Turn Send to the workbench chat off to always type instead. If none is found, check that AI Workbench is running and that its dictation.enabled setting is not off; until then text is typed as usual.
In AI Workbench, the matching setting is dictation.enabled (auto by default). See the workbench settings reference.
Voice corrections follow the text: a correction after dictating into the chat replaces it in the chat.
Share the workbench's model
AI Workbench keeps a small language model running for its own work. MojoFlow can use it for clean-up instead of loading a second one, so your computer holds one model in memory instead of two.
Turn on Clean-up → Mojo Up AI Workbench → Use the workbench's model (on by default; it needs the workbench turned on under Dictation). The Clean-up page says which model is doing the work, and why.
MojoFlow checks it can reach the workbench's model before using it. When the workbench's model stops, is switched off, or runs somewhere MojoFlow cannot reach (a workbench inside WSL, for example), MojoFlow goes back to its own model by itself. The dictation in progress at that moment may be typed without clean-up; it is never lost.