Web Grounding

[laiso]

Improving Summary Accuracy - Fact Verification with Web Grounding

Problem

Currently, Xpaper's summarization feature relies solely on the LLM's internal knowledge. This leads to the following challenges:

1. Lack of Latest Information - Unable to verify recent facts outside the timeline
2. Hallucination - May contain information that differs from facts
3. Absence of Sources - Cannot provide sources to back up information

Solution

Enable Web Search/Grounding features for each LLM provider to allow AI to retrieve latest information and cite sources.

Expected Benefits

• Improved factual accuracy
• Reflection of latest information
• Reliability through source citations
• Reduced hallucinations

Implementation Details

Anthropic (Claude Sonnet 4.6 / Opus 4.6)

Add Web Search tool:

• Tool type: web_search_20250305
• Max searches: 3 per request
• Source display: Markdown list at the end

Gemini

Added googleSearch tool, but verification needed.

OpenAI

Not supported in chat/completions endpoint.

Cost

• Anthropic: $10 / 1,000 searches (separate billing)
• Maximum 3 searches per request