AI Integration Guide

Termboard integrates AI capabilities to accelerate your semantic modeling work. This guide explains the available AI features, how to set them up, and tips for getting the best results.

Overview

Termboard offers AI assistance in two main areas:

1. Per-Term AI - Generate or improve descriptions, find related terms, suggest hierarchy
2. Bulk Generation - Create entire terminology graphs from prompts or text

Both features support multiple AI backends:

Mode	Description	Best For
Cloud AI	Uses cloud API providers (Google, Anthropic, OpenAI, Groq)	High-quality results, fast inference
Local AI	Runs AI models in your browser via WebLLM	Privacy-sensitive work, offline use
Clipboard	Copy prompts to paste into external AI tools	When you don't have API keys

Setting Up AI

Cloud AI Setup

Cloud AI providers offer the best quality and speed but require an API key.

1. Go to Settings > Cloud AI
2. Select your preferred provider
3. Enter your API key
4. Choose a model
5. Click TEST CONNECTION to verify

Recommended Providers:

Provider	Strengths	Free Tier
Google Gemini	Great quality, generous free tier	15 requests/minute
Groq	Very fast inference	Limited free tier
Anthropic Claude	Excellent reasoning	Pay-as-you-go
OpenAI	Well-known, reliable	Pay-as-you-go

Free Option

Google Gemini offers a generous free tier that works well for most users. Get your API key at Google AI Studio.

See Cloud AI Settings for detailed setup instructions.

Local AI Setup (Browser AI)

Run AI models directly in your browser - no API keys, no data leaves your computer.

1. Go to Settings > Local AI
2. Verify WebGPU is supported (you'll see a green message)
3. Select a model size:
   • Small (~1GB) - Fastest, basic quality
   • Medium (~2GB) - Good balance (recommended)
   • Large (~3GB) - Best quality, slower
4. Click Load Model to download and initialize

Browser Requirements

Local AI requires a WebGPU-compatible browser:

• Chrome/Chromium 113+
• Microsoft Edge 113+
• Brave (Chromium-based)

Firefox does not currently support WebGPU.

First-time setup:

• Model downloads are 1-3GB depending on size
• Downloads are cached - subsequent loads are instant
• You can unload models to free memory

See Local AI Settings for troubleshooting.

Per-Term AI Features

Access per-term AI features from the Term Sidebar:

1. Select a term in your graph
2. Click the Search and AI Prompts button (star icon)
3. Choose your AI mode (Cloud AI, Local AI, or Copy Prompt)
4. Select an operation

Available Operations

Operation	What It Does	When to Use
Generate Description	Creates a definition for the term	New terms without descriptions
Improve Description	Enhances an existing description	Refining rough definitions
Generate Explanation	Creates detailed explanation with examples	Documentation, training materials
Suggest Parents	Finds broader/parent terms	Building taxonomies
Suggest Children	Finds narrower/child terms	Expanding taxonomies
Suggest Related Terms	Discovers related concepts and relations	Finding connections

Using AI Suggestions

For Parent, Children, and Related suggestions:

1. AI generates suggestions with descriptions
2. Review the suggested terms and relations
3. Check/uncheck items you want to add
4. Click Import Selected
5. New terms and relations appear in your graph

Context Matters

Provide good descriptions for your terms before requesting suggestions. The AI uses existing context to provide more relevant suggestions.

Bulk AI Generation

For creating larger structures, use the Bulk Add feature:

1. Open Tools > Add in Bulk/Chatbot
2. Select a prompt template from the dropdown
3. Provide required inputs (subject, context, or text)
4. Click USE AI or COPY PROMPT

Popular Prompt Templates

Creating New Models:

• Semantic data model about a subject - Full semantic model with concepts and relations
• Knowledge Graph about a subject - Ontology-style graph
• Mind Map - 20+ concepts in hierarchical structure
• Taxonomy for selected terms - Organize existing terms hierarchically

Extracting from Text:

• Ontology from text - Extract concepts and relations from documents
• Terms, Definitions from Text - Pull terminology from existing content
• User stories from Text - Extract user stories with acceptance criteria

Improving Existing Content:

• Missing Definitions - Generate descriptions for terms without them
• Improve Definitions for selected terms - Enhance existing descriptions
• Relations between selected terms - Find connections in your model

See Add in Bulk/Chatbot for all 28 prompt templates.

Tips for Better Results

1. Be Specific with Context

Instead of just entering a subject like "Banking", add context:

Subject: Banking Context: Focus on retail banking products and services for individual customers, including accounts, loans, and digital banking channels.

2. Use Descriptive Term Names

AI produces better suggestions when terms have:

• Clear, unambiguous names
• Existing descriptions (even brief ones)
• Proper term types (Concept vs. Property)

3. Start with Core Concepts

For large models, work iteratively:

1. Generate core concepts first
2. Review and refine
3. Use "Suggest Children" to expand each area
4. Use "Relations between selected terms" to find connections

4. Leverage Existing Terms

Many prompts can reference existing terms in your graph:

1. Select terms in your graph
2. Use prompts like "Taxonomy for selected terms"
3. AI uses selected terms as context

5. Review and Edit

AI suggestions are a starting point:

• Review generated descriptions for accuracy
• Adjust relation types as needed
• Remove irrelevant suggestions
• Add missing nuances

Privacy Considerations

Cloud AI

When using Cloud AI providers:

• API Keys: Stored locally in your browser's localStorage, never sent to Termboard servers
• Data Transmission: Term names and descriptions are sent to the AI provider's API
• Data Retention: Check each provider's data retention policy
• CORS Proxy: Some providers (OpenAI, Anthropic) require a CORS proxy - requests pass through ai-proxy.m-a-hoogkamer.workers.dev

Sensitive Data

If you're working with confidential terminology, consider using Local AI or the clipboard workflow with an internal AI system.

Local AI (WebLLM)

When using Local AI:

• 100% Private: All processing happens in your browser
• No Network: After initial model download, no internet needed
• No API Keys: No accounts or credentials required
• Browser Storage: Models cached in browser storage (~1-3GB)

This is the best option for:

• Sensitive enterprise terminology
• Air-gapped environments
• Privacy-conscious users

Clipboard Mode

The most private cloud option:

1. Termboard generates a prompt (no data sent anywhere)
2. You paste it into your organization's approved AI tool
3. You copy the response back
4. You review before importing

This gives you complete control over what data leaves your system.

Troubleshooting

Cloud AI Issues

"API Key Invalid"

• Double-check the key was copied correctly
• Ensure the key hasn't expired
• Verify the key has necessary permissions

"Rate Limit Exceeded"

• Wait a moment before retrying
• Google Gemini free tier: 15 requests/minute
• Consider upgrading to a paid tier for heavy use

"Response Truncated"

• Some complex prompts need more tokens
• Try a different model or provider
• Use the "40 Concepts" prompt only with high-capacity models

Local AI Issues

"WebGPU Not Supported"

• Update your browser to Chrome/Edge 113+
• Check chrome://gpu for WebGPU status
• Update GPU drivers

"Model Failed to Load"

• Ensure you have sufficient disk space (~3GB)
• Try a smaller model
• Clear browser cache and retry

Slow Performance

• Local AI is slower than cloud (expected)
• Use a smaller model for faster inference
• Close other tabs to free memory

Comparison: Cloud vs. Local AI

Aspect	Cloud AI	Local AI
Quality	Excellent	Good
Speed	Fast (2-10 sec)	Slower (10-30 sec)
Privacy	Data sent to provider	100% local
Cost	May require paid tier	Free
Offline	Requires internet	Works offline
Setup	API key required	One-time download
Bulk Add	Full support	Not available

Related Resources

• Cloud AI Settings - API configuration
• Local AI Settings - WebLLM setup
• Add in Bulk/Chatbot - Bulk generation
• Term Sidebar - Per-term AI access
• Semantic Data Model - Building models