Skip to content

Web Search & Tools ​

ai.KMITL provides powerful tools that extend the AI's capabilities beyond its training data. This guide covers how to use web search, supermemory, and other integrated tools.

Web search allows the AI to:

  • Search the internet in real-time
  • Access current information
  • Find recent news and updates
  • Get data beyond its training cutoff date

Web Search Enabled

To enable Web Search, simply click the Globe icon located in the chat toolbar below the message input box.

  • When web search is enabled, the Globe icon will appear highlighted (see image above).
  • If the button appears gray, web search is currently disabled.

Enabling web search allows the AI to access up-to-date information and answer questions about recent news, current events, or anything beyond its training data.
You can toggle this feature on or off at any time before sending your message.

Use web search for:

  • ✅ Current events: "What happened in the news today?"
  • ✅ Latest information: "Is "เสนาหอย" play as Superman?"
  • ✅ Real-time data: "What's the weather in Bangkok?"
  • ✅ Recent releases: "When is the next iPhone coming out?"
  • ✅ Product information: "Best laptops under 30000 baht"
  • ✅ Recent updates: "Latest features in ChatGPT"

Don't need web search for:

  • ❌ General knowledge: "What is photosynthesis?"
  • ❌ Historical facts: "When did World War 2 end?"
  • ❌ Programming help: "How to write a for loop in Python?"
  • ❌ Math problems: "Solve this equation"
  • ❌ Creative writing: "Write me a story"
  • ❌ Analysis of uploaded files: The AI already has the file

How Web Search Works ​

When web search is enabled:

  1. You send a message
  2. AI determines if search is needed
  3. Searches relevant websites (if needed)
  4. Reads the top result pages when snippets aren't enough, and a small, fast model condenses each page into short notes with its title and link
  5. Synthesizes information from those notes
  6. Responds with cited sources

Condensing the pages keeps responses fast and cheap: the AI reads a page or two of focused notes per search instead of whole web pages. Open a search in the chat and expand Summary to see the notes the AI worked from.

Search Limit ​

Each response can run up to 3 web searches (the Globe button's tooltip shows this). The AI plans its searches around the limit; if it tries a fourth, it answers with what it already found. For questions that need many sources, it uses Deep Research instead.

Search Results ​

The AI will:

  • Cite sources with links
  • Summarize findings from multiple pages
  • Compare information across sources
  • Provide recent data when available

Example response:

According to Spin9arm <Link to Spin9arm>, Apple announced new iPhone 99 Pro Max...

Search Providers ​

ai.KMITL uses multiple search providers:

  • TinyFish - Free, agent-oriented search with page extraction
  • Brave Search - Privacy-focused (BYOK)
  • Tavily - AI-optimized search
  • Serper - Google-powered results

The system uses TinyFish as the default provider. Other providers require your own API key.

Deep Research ​

What is Deep Research? ​

For broad questions — comparing products, reviewing a market, collecting everything known about a topic — the AI can hand the job to Deep Research. It is available whenever web search is on; you don't need to turn anything else on.

Deep Research runs on a cheaper research model that works step by step:

  1. Breaks your question into the sub-questions that matter
  2. Runs focused searches (and reads key pages such as official docs or pricing pages)
  3. Writes a cited report

The AI then answers you from that report. In the chat you'll see a Deep research card with live progress (searches and pages so far); once it finishes, expand it to see every search, the sources, and the full report.

Limits ​

Each run stays within fixed limits so it finishes in a few minutes and doesn't drain your quota:

LimitPer run
Runs1 per response
Searches12
Page reads8
Timeabout 6 minutes
Spendup to about $0.50 of quota, and never more than you have left

When a limit is reached, the research model stops searching and writes its report from what it has; the card notes why it stopped early.

Supermemory ​

What is Supermemory? ​

Supermemory is a knowledge base that:

  • Stores information across conversations
  • Remembers important facts
  • Recalls previous discussions
  • Provides long-term context

Coming Soon

Supermemory integration is currently in development and will be available soon!

How It Works ​

Once available, Supermemory will:

  1. Store key information you share
  2. Index important facts
  3. Recall relevant memories when needed
  4. Build a personal knowledge base

Use Cases ​

  • Remember your preferences
  • Recall project details
  • Store frequently used information
  • Build a personal wiki

Image Generation ​

What is Image Generation? ​

With image generation on, the AI can create and edit images for you in the chat. It uses the same engine as Image Studio (Nano Banana 2 Lite):

  • Create pictures, illustrations, logos, and posters from a description
  • Edit an image you uploaded, or one it made earlier ("make the background blue")
  • Combine up to 4 reference images

Enabling Image Generation ​

  1. Click the Tools and modes button (the sliders icon next to the Globe)
  2. Turn on Image generation

The switch is available only for models that support images (multimodal models) and tools. It is paused while Economy Mode is on.

Where Images Go ​

  • Each image appears in the chat, with a Download button
  • It is saved to the chat's Files panel (for example images/poster.png). Click the image to open it there
  • It is also added to your Image Studio collection
  • When you ask for changes, the AI saves the new image as a new version of the same file, so earlier versions stay available

Quota

Each image uses your usage quota, separately from the chat message itself. The AI generates one image per request unless you ask for more.

The AI cannot see the result

The AI writes the prompt but does not see the generated image. If something looks wrong, say what to change.

Questions & Quizzes ​

How the AI Asks You Questions ​

When a request is unclear, the AI can show you a short form instead of guessing. It is always available on models that support tools; there is nothing to turn on. Each question can be:

  • Single choice: pick one option
  • Multiple choice: pick every option that applies
  • Written answer: type your answer in a text box

Clarifying questions also include an Other choice, so you can write your own answer when none of the options fit. Use Skip on any question you don't want to answer, then Submit. You can also ignore the form and reply in the chat as usual.

Quizzes and Practice ​

Ask the AI to quiz you ("Quiz me on HTTP status codes") and it sends the questions as a form. After you submit, it grades each answer, explains your mistakes, and gives your score. Skipped questions count as unanswered. With Tutor mode on, the AI also checks your understanding with short questions while it teaches.

After you submit, the form keeps your choices highlighted, so the chat doubles as a record of the quiz.

MCP (Model Context Protocol) ​

What is MCP? ​

MCP allows integration with external tools and services:

  • File systems - Access local files
  • Databases - Query data sources
  • APIs - Connect to web services
  • Custom tools - Add specialized functions

Advanced Feature

MCP is designed for advanced users and requires additional setup.

Available MCP Servers ​

When enabled, you might have access to:

  • File system access
  • Database connections
  • API integrations
  • Custom tools specific to your needs

Enabling MCP Tools ​

  1. Go to Settings → AI Options → MCP Servers
  2. Configure MCP Servers
  3. Enable specific tools

Tool Usage Tips ​

Best Practices ​

For Web Search:

  1. Be specific about what you're looking for
  2. Mention dates if time-sensitive
  3. Ask for sources if you need citations
  4. Compare results for accuracy

General Tips:

  1. Enable tools only when needed to save processing time
  2. Disable tools for general conversations
  3. Check sources provided by web search
  4. Verify critical information from multiple sources

Example Queries ​

Good Web Search Queries:

✅ "What are the latest AI developments announced this week?"
✅ "Find current pricing for OpenAI API"
✅ "What's the weather forecast for Bangkok tomorrow?"
✅ "Latest news about KMITL"

Queries That Don't Need Web Search:

❌ "Explain machine learning"
❌ "Help me write Python code"
❌ "Solve this math problem"
❌ "Summarize this uploaded document"

Tool Indicators ​

Active Tools ​

When tools are active, you'll see:

  • Web Search search results in chat area
  • Tool icon and status indicator in the response
  • Source citations in the response
  • Search queries used in the response

Privacy & Data ​

Web Search Privacy ​

  • Searches are proxied through secure servers
  • Your IP address is not exposed to target sites
  • Search history is not tracked externally
  • Results are not stored permanently

Tool Data Handling ​

  • Tool interactions are encrypted
  • Data is processed securely

Troubleshooting ​

Web Search Not Working? ​

Check these:

  1. Is the Globe (web search) button highlighted?
  2. Try rephrasing your query
  3. Check if the service is temporarily down

Slow Responses with Tools? ​

Web search adds time because:

  • Searches multiple websites
  • Reads and processes results
  • Synthesizes information

Tips for speed:

  • Disable tools for non-time-sensitive queries
  • Be specific to reduce search scope
  • Use faster AI models

Inaccurate Results? ​

If web search gives wrong info:

  1. Check the sources provided
  2. Try a different query
  3. Verify with multiple searches
  4. Report issues to support

Tool Limitations ​

Web Search Limitations ​

  • Cannot access paywalled content
  • May miss very recent updates (days old)
  • Depends on search provider availability
  • Some sites block automated or AI access

General Limitations ​

  • Tools add processing time
  • May occasionally fail
  • Not available for all models
  • Some features require BYOK

Tool Costs ​

Included Quota ​

Tool work done by the platform's own models counts toward your usage quota:

  • Page notes for web search (a small model condenses the pages it reads) cost a fraction of a cent per search.
  • Deep Research charges what its research model and page notes actually cost, capped per run (see Limits).

These appear in Settings → Usage under the models that did the work.

BYOK ​

With your own API keys:

  • Unlimited tool usage
  • All features available
  • You pay only for external API costs
  • See BYOK guide

Experiment with Tools

Try enabling and disabling tools to see how they affect responses. You'll quickly learn when each tool is most useful!

Next Steps ​

Made with ❤️ by KDMC (KMITL Data Management Center)