Prompt Management
Prompts are the instructions you give to DuRT's AI (LLM) services. Instead of typing an instruction every time, you save prompts once and reuse them across sessions. DuRT organizes prompts into five categories so you always have the right instruction ready for the task at hand.
What Is a Prompt?
When DuRT sends transcribed text to a large language model (LLM), it includes a prompt — a short instruction that tells the AI what to do with the text. For example:
- "Translate the following text into Chinese."
- "Summarize the key points in three bullet points."
- "Fix any grammar or spelling errors without changing the meaning."
Prompts are stored in DuRT so you can select them quickly inside your workflow without retyping them each time.
The Five Prompt Categories
DuRT groups prompts into five categories, each designed for a specific role in your AI workflow:
1. Sentence Splitting
Prompts in this category instruct the AI to break long, continuous ASR output into properly separated sentences or paragraphs. Raw transcription often lacks punctuation and natural sentence boundaries — a sentence splitting prompt cleans this up before further processing.
Example use: After transcribing a 10-minute meeting, apply a sentence splitting prompt to turn the wall of text into readable, punctuated sentences.
2. Translation
Translation prompts instruct the AI to convert text from one language to another. You can create multiple translation prompts for different language pairs.
Example use: Translate English podcast transcripts into Spanish, or convert Japanese lecture notes into English.
3. Proofreading
Proofreading prompts ask the AI to correct grammar, fix spelling mistakes, and improve sentence clarity — without changing the original meaning or tone.
Example use: Clean up the raw output of an ASR engine that occasionally mishears words, producing a polished, readable transcript.
4. Summary
Summary prompts instruct the AI to condense a long transcription into a concise overview, highlight key points, or generate meeting minutes.
Example use: After a one-hour team meeting, generate a short bullet-point summary of decisions and action items.
5. Custom Prompt (Other)
The "Other" category is a catch-all for any prompt that does not fit the above categories. Use it for specialized tasks like keyword extraction, tone adjustment, formatting output as a table, and more.
Built-in vs. Custom Prompts
| Type | Description |
|---|---|
| Built-in Prompts | Included with DuRT by default. They cover common tasks and are a good starting point. They cannot be deleted, but you can duplicate and modify them. |
| Custom Prompts | Prompts you create yourself. Fully editable, renameable, and deletable. |
TIP
Start with a built-in prompt to see what works, then duplicate it and fine-tune the wording to match your specific workflow.
Creating a New Prompt
- Open Settings and go to the Prompt tab.
- Click the ＋ Add button.
- Choose a category (Sentence Splitting, Translation, Proofreading, Summary, or Other).
- Enter a name for the prompt (e.g., "EN → ZH Translation").
- Write the prompt text in the content field. Be specific about what you want the AI to do.
- Click Save.
Your prompt is now available in any LLM workflow step.
NOTE
The quality of AI output depends heavily on how clearly you write your prompt. Include details like target language, desired format, and tone when relevant.
Using a Prompt in an LLM Workflow Step
DuRT's workflow engine processes your transcription through a series of steps. Each LLM step can be assigned a prompt:
- Open your workflow configuration in DuRT.
- Find the LLMProcess step (or add one if it does not exist).
- Click the Prompt selector for that step.
- Choose the prompt you want from the dropdown list, filtered by category.
- Save the workflow.
From now on, every time that workflow runs, the selected prompt is automatically sent to the LLM along with your transcribed text.
IMPORTANT
You can chain multiple LLM steps together — for example, first a Sentence Splitting step, then a Translation step. Each step uses its own independently selected prompt.
Practical Examples
Podcast Translation
Workflow: ASR → Sentence Splitting → Translation
Set the Sentence Splitting step to a prompt that adds punctuation, then set the Translation step to a prompt like "Translate the following English text into Chinese. Keep a natural, conversational tone."
Meeting Minutes
Workflow: ASR → Proofreading → Summary
The Proofreading step cleans up recognition errors, and the Summary step generates structured meeting notes with key decisions and next steps.
Language Learning Review
Workflow: ASR → Translation → Custom (Vocabulary Highlight)
Translate a foreign-language lecture and then apply a custom prompt to extract unfamiliar vocabulary and provide brief definitions.
Tips for Writing Effective Prompts
- Be specific: Instead of "translate this," say "translate this English text into formal Traditional Chinese."
- Define the output format: If you want bullet points, say so. If you want a single paragraph, specify that.
- Set the tone: "Professional and concise" yields different results from "casual and friendly."
- Iterate: If the output is not what you expected, tweak one part of the prompt at a time and test again.