Skip to content

User Guide

InspectChat (Chat Interface)

InspectChat is LLMInspect's primary interface for interacting with AI models. It provides a familiar chat-like experience while maintaining enterprise-grade security and monitoring.

GUI Overview

Main Interface Components

The InspectChat interface consists of five key areas designed for intuitive interaction:

Main Interface

Figure: InspectChat Main Interface Layout

  1. Chat Window: The central conversation area where messages and responses appear

  2. Message Input Field: Locate-

d at the bottom for typing your prompts

  1. Model Selector: Top menu dropdown for choosing AI models

  2. Settings Menu: Gear icon for accessing configuration options

  3. History Panel: Left sidebar for viewing conversation history

  4. File Management: Right sidebar for managing uploaded documents

  • Use keyboard shortcuts: Enter to send messages, Shift+Enter for new lines

  • Scroll through chat history using mouse wheel or scroll bar

  • Conversations are automatically saved and accessible via the History Panel

Model Selection

Available AI Models

InspectChat provides access to multiple AI providers and specialized models:

OpenAI Models

  • GPT-4o and GPT-4o-mini

  • o3 and o4-mini

Google Models

  • Gemini 1.5-pro

Eunomatix Models

  • InspectGPT: Our fine-tuned model for enterprise use

Plugin-Based Models

  • DALL-E 3: For image generation

  • Additional plugins available in the plugin store

How to Select a Model

Model Selection Interface

Figure: Model Selection Interface

  1. Access the selector: Click the model dropdown in the top menu bar

  2. Browse options: Scroll through available models organized by provider

  3. Select your model: Click on your preferred model

  4. Confirm selection: The interface updates to show your active model

  5. Model-specific features: Some models may show additional options or capabilities

Text Generation

Basic Chat Interaction

Text Generation Example

Figure: Text generation example

Starting a conversation:

  1. Select your model using the model selector

  2. Type your message in the input field at the bottom

  3. Send your message by pressing Enter or clicking the Send button

  4. Wait for response - processing time varies by model and complexity

  5. Continue the conversation - each exchange builds context

Message Input Tips:

  • Single line: Press Enter to send

  • Multiple lines: Use Shift+Enter to create line breaks

  • Edit before sending: Click in the text area to modify your message

  • Clear input: Use Ctrl+A then Delete to clear the field

Advanced Text Generation Features

Context Management:

  • Conversations maintain context across multiple exchanges

  • Earlier messages influence later responses

  • Long conversations may be truncated based on model limits

Response Controls:

  • Stop generation early if needed

  • Regenerate responses for different outputs

  • Copy responses using the copy button

Best Practices for Text Generation

Prompt Engineering:

  • Be specific: "Write a Python function to sort a list" vs "Help with code"

  • Provide context: Include relevant background information

  • Use examples: Show the format or style you want

  • Break down complex tasks: One request per message for better results

Effective Communication:

  • Keep messages clear and concise

  • Use proper grammar and spelling

  • Specify the desired output format

  • Ask follow-up questions to refine responses

Common Patterns: 

Good: "Explain quantum computing in simple terms for a business audience"
Better: "I need to explain quantum computing to my executive team who have no technical background. Please provide a 2-minute explanation focusing on business applications and potential benefits."

Image Generation

Setting Up Image Generation

Image generation in InspectChat is powered by plugins. Here's how to get started:

DALL-E 3 Plugin Selection

Figure: Choose DALL-E 3 and install it

Installation Steps:

  1. Select Plugin Provider: Choose "Plugins" from the LLM Provider dropdown

  2. Access Plugin Store: The interface will show available plugins

  3. Find DALL-E 3: Locate the DALL-E 3 plugin in the store

  4. Install Plugin: Click the install button

  5. Confirm Installation: Wait for the plugin to be activated

Creating Images

Image Generation Example

Figure: Example use case of Image generation

Image Generation Process:

  1. Ensure DALL-E 3 is selected as your active model

  2. Write your image description in the message input field

  3. Be descriptive: Include style, colors, composition, mood

  4. Send your request: Click "Send Message"

  5. Wait for generation: Image creation typically takes 10-30 seconds

  6. View and download: The generated image appears in the chat

Writing Effective Image Prompts

Prompt Structure: 

[Subject] + [Style] + [Setting] + [Mood/Atmosphere] + [Technical details]

Examples:

  • Basic: "A red car"

  • Better: "A sleek red sports car parked in front of a modern building"

  • Best: "A glossy red Ferrari 488 GTB parked in front of a glass skyscraper during golden hour, cinematic lighting, professional photography style"

Style Keywords:

  • Artistic: "watercolor", "oil painting", "digital art", "sketch"

  • Photography: "portrait", "landscape", "macro", "street photography"

  • Mood: "dramatic", "peaceful", "energetic", "mysterious"

Technical Specifications:

  • Quality: "high resolution", "detailed", "sharp"

  • Composition: "close-up", "wide angle", "bird's eye view"

  • Lighting: "soft lighting", "dramatic shadows", "bright and airy"

Document Upload & Processing

Supported File Types

InspectChat can process various document formats for analysis and question-answering:

File Type Extension Max Size Best For
PDF .pdf 5MB Reports, manuals, research papers
Text .txt 5MB Plain text documents, logs
Word .doc, .docx 5MB Business documents, proposals
Excel .xlsx 5MB Data analysis, spreadsheets

Upload Process

Method 1: Drag and Drop

  1. Select your file in your file explorer

  2. Drag the file into the InspectChat window

  3. Drop it anywhere in the chat area

  4. Wait for processing - you'll see a processing indicator

  5. Start asking questions once processing is complete

Method 2: File Selection (if available)

  1. Click the attachment icon (if present in your interface)

  2. Browse and select your file

  3. Confirm upload

  4. Wait for processing

Document Processing Features

What happens during processing:

  • Text extraction from various formats

  • Content indexing for quick retrieval

  • Metadata extraction (title, author, creation date)

  • Security scanning (if GuardRails are enabled)

Processing indicators:

  • Upload progress bar

  • "Processing document..." message

  • Completion notification

  • File appears in the Manage Files panel

Working with Uploaded Documents

Asking Questions:

Once processed, you can interact with your documents:

Examples:

- "Summarize the main points in this document"

- "What does the document say about quarterly revenue?"

- "Find all mentions of 'machine learning' in the uploaded file"

- "Compare the data in columns A and B of the spreadsheet"

Document Management:

  • View uploaded files: Check the "Manage Files" option in the right sidebar

  • Remove documents: Delete files you no longer need

  • Multiple documents: Upload several files for comparative analysis

  • Document persistence: Files remain available throughout your session

Best Practices for Document Processing

Before Uploading:

  • Ensure documents are text-readable (not scanned images without OCR)

  • Remove sensitive information if not needed for analysis

  • Check file size stays under 5MB limit

  • Use clear, descriptive filenames

Optimizing Questions:

  • Reference specific sections: "In chapter 3, what..."

  • Ask for specific formats: "Create a bullet-point summary of..."

  • Request comparisons: "How does this differ from..."

  • Seek clarifications: "Explain the technical terms mentioned in..."

Troubleshooting Upload Issues:

  • File too large: Compress or split the document

  • Unsupported format: Convert to PDF or TXT

  • Processing failed: Check if the document is corrupted

  • No text extracted: Document might be image-based