Veriprompt Docs

Appearance

Web Search Augmentation ​

VeriPrompt can automatically enrich AI responses with live web data when your question needs real-time information. This feature uses Tavily as a search backend and works transparently -- no changes to how you chat.

How It Works ​

  1. You send a message in the chat terminal (e.g., "What's the weather in Berlin?")
  2. VeriPrompt detects that the question needs live data
  3. A web search runs in the background (under 5 seconds)
  4. Search results are injected into the AI's context
  5. The AI answers using both its training data and fresh web results

The search results are never shown to you directly -- instead, the AI uses them to provide an accurate, up-to-date answer and may cite sources inline.

Web search activates when your message contains time-sensitive keywords:

CategoryExample Triggers
Weather"weather", "forecast"
News"news", "headlines", "breaking"
Recency"latest", "current", "recent", "trending"
Time"today", "yesterday", "tomorrow", "this week"
Finance"stock price", "crypto", "bitcoin", "market"
Sports"score", "standings", "who won"
Events"release date", "when does", "when will"

Messages that don't match any trigger -- like "Write me a poem" or "Explain JavaScript closures" -- skip the search entirely and go straight to the AI.

Examples ​

You:  What's the weather in Leipzig tomorrow?
AI:   Tomorrow in Leipzig, expect partly cloudy skies with a high of 12C
      and a low of 4C. Light winds from the northwest at 15 km/h.
      Source: weather.com
You:  What are the latest AI news headlines?
AI:   Here are today's top AI stories:
      1. OpenAI announces GPT-5 preview...
      2. EU AI Act enforcement begins...
      (Sources: techcrunch.com, reuters.com)

Without Web Search (No Trigger) ​

You:  Write a haiku about coding
AI:   Lines of logic flow
      Bugs hide in the whitespace gaps
      Coffee fuels the fix

Setup (Admin) ​

Web search is optional. To enable it:

  1. Sign up for a free API key at tavily.com
  2. Add the key to your environment configuration:
    bash
    TAVILY_API_KEY=tvly-your-key-here
  3. Restart the application

No database changes or UI configuration required.

Checking Status ​

  • Enabled: Server logs show [WEB_SEARCH] Augmented with Tavily results for qualifying queries
  • Disabled: No TAVILY_API_KEY set -- chat works normally without any search features

Graceful Degradation ​

Web search is designed to never break your chat experience:

ScenarioBehavior
No TAVILY_API_KEY setSearch is completely disabled; chat works normally
Invalid or expired API keySearch silently fails; AI answers from training data only
Tavily API is downSearch times out after 5 seconds; AI answers normally
Rate limit exceeded (429)Search skipped for that request; logged as warning
Non-search querySearch is never triggered; zero overhead

In all failure cases, the chat continues without interruption. Warnings are logged server-side for admin visibility but are never shown to end users.

Troubleshooting ​

AI still says "I can't browse the internet" ​

  • Verify TAVILY_API_KEY is set and valid
  • Check server logs for [WEB_SEARCH] entries
  • Ensure the query contains a trigger keyword (see table above)

Responses feel slow on some queries ​

  • Web search adds up to 5 seconds on triggered queries
  • Non-triggered queries have zero additional latency
  • If Tavily is consistently slow, check your plan's rate limits

Search results seem outdated ​

  • Tavily returns results based on its index freshness
  • For time-critical data, try more specific queries (e.g., "Bitcoin price USD right now")