Fast Mode
Instant search metadata for lightning-fast filtering and intelligent content selection in LLM-powered workflows.
Fast Mode is an optimized search option that returns search result metadata (titles, snippets, URLs, positions) without fetching and parsing the full page content. This mode is designed for speed and efficiency when you need quick access to search results for analysis, filtering, or selective content extraction.
Overview
By default, the Search API fetches and parses the full content of each search result. Fast Mode bypasses the content extraction step, delivering search result metadata (title, snippet, URL, position) fields in a lightweight, fast response.
When to Use Fast Mode
Fast Mode is ideal for several use cases:
Intelligent Content Selection Analyze titles and snippets to determine which sources are worth extracting. Feed metadata to your LLM or filtering logic, then selectively fetch full content only from relevant results using the /extract endpoint. Perfect for agentic workflows and two-step pipelines.
High-Volume Operations Reduce latency and resource usage when performing large-scale search operations. Get metadata for hundreds of queries without the overhead of fetching full page content.
Search Monitoring & Analytics Track what content appears in search results over time for competitive analysis, SEO monitoring, or research purposes - without the cost of extracting full pages.
Usage
Enabling Fast Mode
To enable Fast Mode, set deep_search to false in your search request:
curl -X POST https://nimble-retriever.webit.live/search \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "latest AI trends 2025",
"num_results": 10,
"deep_search": false,
"country": "US",
"locale": "en"
}'Request Parameters
All standard search parameters are supported in Fast Mode:
query
string
Yes
-
The search query to execute
num_results
integer
No
3
Number of URLs to return (max: 100)
deep_search
boolean
Yes
-
Must be set to false for Fast Mode
locale
string
No
en
Locale for search results
country
string
No
US
Country code for search results
Note: The parsing_type, render, driver, and wait parameters are ignored in Fast Mode since no content extraction occurs.
Response Format
Fast Mode Response
The response includes search result metadata with empty page_content fields:
{
"message": "Request processed successfully",
"body": [
{
"page_content": "",
"metadata": {
"title": "Latest GenAI Trends: 5 Key Developments to Watch",
"snippet": "Explore the top trends in generative AI including agentic AI, retrieval-augmented generation, self-training models, and ethical AI implementations.",
"url": "https://example.com/genai-trends-2025",
"position": 1,
"entity_type": "OrganicResult",
"country": "US",
"locale": "en"
}
},
{
"page_content": "",
"metadata": {
"title": "GenAI Investment and Market Analysis 2025",
"snippet": "Despite challenges, investment in generative AI continues to grow. Analysis of market trends, adoption rates, and future projections for enterprise AI.",
"url": "https://example.com/genai-investment-analysis",
"position": 2,
"entity_type": "OrganicResult",
"country": "US",
"locale": "en"
}
},
{
"page_content": "",
"metadata": {
"title": "Emerging Technologies: AI and Data Trends",
"snippet": "Comprehensive overview of agent-ready data, AI for data engineering, data provenance, compliance changes, and the rise of agentic edge computing.",
"url": "https://example.com/emerging-tech-trends",
"position": 3,
"entity_type": "OrganicResult",
"country": "US",
"locale": "en"
}
}
]
}Each result includes:
title: The page title from search results
snippet: A brief description or excerpt from the search result
url: The URL of the result
position: The ranking position in search results
country and locale: Search parameters used
Performance Benefits
3-5x faster response times compared to standard mode
10-100x smaller response payloads for easier processing
Lower resource consumption for high-volume operations
Reduced bandwidth usage for network-constrained environments
Combining with Extract Endpoint
Fast Mode pairs perfectly with the /extract endpoint for efficient two-step workflows:
Discovery Phase: Use Fast Mode to find all relevant URLs
Filter Phase: Apply your own logic to select which URLs to process
Extraction Phase: Use
/extractto get full content from selected URLs
This approach gives you full control over which content to fetch while maintaining high performance.
LLM & AI Application Benefits
Fast Mode is particularly valuable for LLM and AI-powered applications:
Efficient Token Usage Metadata (title + snippet) uses significantly fewer tokens than full page content - often 10-20x less. This keeps your context window available for reasoning and task execution rather than raw content, while dramatically reducing API costs for token-based LLM services.
Intelligent Source Selection Provide your LLM with enough context (titles and snippets) to make informed decisions about source relevance without reading full content. Enable agentic workflows that analyze metadata first, then selectively extract only the most promising sources.
Scalable Search Operations Perform multiple search iterations to explore different angles, refine queries, and analyze results across topics without exhausting your context window or budget. Perfect for research agents and multi-step workflows.
Example Workflows
Two-Step Content Extraction
# Step 1: Fast Mode - Get metadata
curl -X POST https://nimble-retriever.webit.live/search \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "AI research papers 2025",
"num_results": 20,
"deep_search": false
}'
# Response includes title, snippet, URL for each result
# Step 2: Filter based on metadata (e.g., only .edu domains or specific keywords)
# Step 3: Extract - Fetch specific URLs
curl -X POST https://nimble-retriever.webit.live/extract \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"links": [
"https://university.edu/paper1",
"https://research-lab.edu/paper5"
],
"parsing_type": "markdown"
}'Last updated