Utility Modes

Utility Modes

The Modes tab configures the underlying artificial intelligence logic for dIKta.me's fixed utility pipelines.

While the "Dictation Modes" tab lets you create infinite customizable profiles to switch between, the single Modes tab directly locks in the behavior for your core hotkeys: Ask, Refine, Translate, and Note.

Pipeline Configuration

Every pipeline has a dedicated sub-section on this screen. Selecting a pipeline from the list will open its specific Cloud and Local system prompt editors (exactly like a standard Dictation preset).

  1. Ask Pipeline: The prompt sent to the LLM when you use Ctrl+Alt+A.
    • Default: "You are a helpful assistant. Provide concise, direct answers to the following question. Do not include conversational filler like 'Here is the answer'. Just answer the user."
  2. Refine Pipeline: The prompt sent to the LLM when you highlight text and press Ctrl+Alt+R (specifically in Autopilot mode).
    • Default: "Fix any spelling or grammatical errors in the following text. Do not change the underlying tone or meaning. Maintain formatting."
  3. Note Pipeline: The prompt sent when you dictate a Ctrl+Alt+N diary entry.
    • Default: "Format the following transcript into a coherent, properly punctuated note. Fix spelling mistakes but do not omit any information."
    • Note: If you have "LLM Processing" disabled in the General settings, these prompts are ignored entirely.
  4. Translate Pipeline: The prompt sent when translating text natively via Ctrl+Alt+T.
    • Default: "Translate the following text into English. Output only the translated text, nothing else."

Context Injection

Some pipelines rely on injecting dynamic information into the model's instructions right before execution:

  • Refine ({instruction}): If you use Voice Instruction mode for Refine, dIKta.me will replace {instruction} in your written prompt with whatever you actually said out loud. Ensure your prompt includes {instruction} so the model knows what to do with the audio command!

Cloud vs Local Tuning

As always, dIKta.me ensures your pipelines are robust whether you are online or entirely offline.

You must supply both a Cloud System Prompt (for models like gpt-4o or gemini-1.5-pro) and a Local System Prompt (for local Ollama models like llama3 or phi3) for every utility mode to guarantee success when you flip the Control Panel Environment switch.