Features

Features improve user interaction with digital experts, e.g., adding file upload, self-introduction, or welcome messages for richer experiences.

Click the “Features” button in the app’s top-right corner to add enhancements.

Features Details

1. Welcome Message

Custom text on the welcome page, introducing the AI or guiding users.

Purpose:

• Build rapport
• Encourage questions
• Clarify app context

Setup:

• Customizable message (supports line breaks, emojis)
• 1–10 suggested questions as clickable buttons

Tips:

• Keep it concise, e.g., “Hi, I’m your data analysis assistant for sales, inventory, or budgets.”
• Suggested questions: “View this month’s sales,” “Analyze customer churn,” “Generate financial summary.”

2. Suggested Questions

Generates 5 related follow-up questions after each conversation turn to sustain engagement.

Purpose:

• Maintain user interest
• Lower question barrier
• Spark creative topics

Setup:

• Edit prompt template for question style (professional, casual, humorous)
• Include scenario or instructions

Tips:

• Tailor prompts to app goals, e.g., “Generate 5 humorous, imaginative questions to encourage exploration.”
• Example outputs:
  • What’s an AI’s dream job?
  • Why’s the “F” key cooler than “J”?
  • What’s a cat’s first law if it ruled humans?
  • Biggest time travel bug?
  • What if Monday vanished?

3. Conversation Summary

Summarizes key info from long conversations (e.g., >100 messages) into a concise summary to reduce system load and improve focus.

Mechanism:

• Max Tolerance: Max new messages before summarizing (e.g., 22).
• Messages to Summarize: Number of messages summarized (e.g., 16 of 22).
• Retained Messages: Messages kept post-summary (e.g., 4; rest deleted, stored as summary in system prompt).
• Prompt: Guides summary; defaults if unset.

Example Config:

• Max tolerance: 100
• Summary trigger: 40
• Retained: 60 messages

Use Cases:

• Long interactions
• Multi-turn task tracking
• Periodic knowledge summaries

4. Long-term Memory

See Long-term Memory.

5. Memory Reply

Manually store Q&A pairs for consistent, controlled replies, bypassing LLM generation. Ideal for standardized or hallucination-sensitive scenarios.

Mechanism:

• Q&A Creation: Developers write question-answer pairs as long-term memory.
• Similarity Matching: System matches user input to stored questions via Embedding vectors.
• Score Threshold: Returns stored answer if similarity ≥ threshold.
• Embedding Model: Uses long-term memory’s model or system default.

Parameters:

• Score Threshold: Similarity limit (e.g., 0.85).
• Embedding Model: Configured model.

Use Cases:

• Standardized replies (e.g., “What’s the customer service number?”)
• Rapid prototyping (POC/DEMO)
• Stable fallback with RAG

Advantages:

• Consistent, predictable answers
• Reduces LLM hallucinations
• Works alongside knowledge bases, search

6. Conversation Title

Auto-generates concise titles for new conversations based on content, improving browsability.

Config:

• Enabled by default
• Customizable prompt for title style/format

Examples:

• “Sales department last month’s performance” → Sales Performance Analysis
• “Write a casual AI assistant intro” → AI Assistant Intro Request
• “Explain memory reply mechanism” → Memory Reply Mechanism Explained

Tips:

• Use prompts for concise, specific titles, e.g., “Summarize the query’s theme in one sentence, avoiding redundancy.”

7. File Upload

The file upload function allows users to upload specific file types during a conversation, inserting images or full text into the current dialogue to help the AI better understand and respond to contextual questions.

✅ Function Description

• Upload Location: Users can upload files by clicking the 📎 icon (or “Upload a file” button) in the dialogue box.

• Post-Upload Effect: File content (e.g., document text, images) is injected into the context as user messages for the model to reference, improving dialogue comprehension accuracy.

⚙️ Configuration Options

• 1. Maximum Upload Quantity
  • Value Range: 1 to any value
  • Meaning: Limits the number of files a user can upload at once
  • Example: Maximum 10 files
• 2. File Size Limits (by Type)

File Type	Maximum Single File Size
Document	15MB
Image	10MB
Audio	50MB
Video	100MB

• 3. Supported File Types

The following file types are supported for upload, facilitating large model comprehension and parsing:

• Documents: TXT, MD, MDX, MARKDOWN, PDF, HTML, XLSX, XLS, DOC, DOCX, CSV, EML, MSG, PPTX, PPT, XML, EPUB
• Images: JPG, JPEG, PNG, GIF, WEBP, SVG
• Audio (if enabled): MP3, M4A, WAV, AMR, MPGA
• Video (if enabled): MP4, MOV, MPEG, WEBM

Supported file types can be configured via interface checkboxes; non-selected types will be prohibited from upload.

8. Text to Speech (TTS)

Converts AI responses to audio via a play button, enhancing accessibility.

Config:

• TTS Model: Azure, Google, Edge-tts, or local.
• Parameters: Voice settings.
• Trigger: Play/stop via button.
• Multilingual: Adapts to supported languages.

9. Speech to Text (STT)

Converts voice input to text via microphone button.

Config:

• Activation: Click microphone to record.
• Browser Permissions: Requires mic access.
• STT Model: Web Speech API or external (Whisper, Azure).
• Language: Preset for accuracy.

Privacy:

• Voice used only for transcription, not stored (unless authorized).
• Mic permission can be revoked.

FAQ

1. Feature not working?

Causes/Solutions:

• Unsaved Config: Save settings.
• Incomplete Config: Check for missing model/parameters.
• Network/Service Issue: Verify connection, check console for API errors.
• Suggested Questions: Only active post-release.

2. File upload fails?

Issues/Solutions:

• Size Limit: Check file size (e.g., ≤20MB).
• File Type: Supports .pdf, .docx, .xlsx, .csv, .txt.
• Network/Server: Check console for API errors.
• Browser: Use Chrome, Edge, Safari.
• Context Binding: Ensure file is linked to conversation.

3. Clear long-term memory?

Clear All:

1. Go to memory management.
2. Click “Clear Memory” and confirm.
3. Deletes all memory entries (irreversible).

Delete Specific Entries:

1. In memory management, select entries.
2. Click “Delete” or search and delete.

Note:

• Doesn’t affect short-term history.
• Clearing removes customization; system reverts to base model.
• Operation is irreversible.