UnSearch Python SDK

Comprehensive AI search API with full backend capabilities. Drop-in Tavily replacement with self-hosting and zero-retention options.

Features

• Tavily-compatible - Drop-in replacement for Tavily API
• Exa-compatible - Neural/semantic search endpoints
• RAG-optimized - Built for retrieval augmented generation
• 70+ search engines - Aggregated results from multiple sources
• Topic monitoring - Real-time web monitoring
• Fact verification - AI-powered fact-checking
• Zero-retention - Privacy-first architecture

Installation

pip install unsearch

# With LangChain support
pip install unsearch[langchain]

Quick Start

from unsearch import UnSearchClient

client = UnSearchClient(api_key="your-api-key")

# Basic search
response = client.search("What is machine learning?")
for result in response.results:
    print(f"{result.title}: {result.url}")

# Search with AI answer
response = client.search(
    "What is RAG?",
    include_answer=True,
    max_results=5
)
print(response.answer)

# Q&A shortcut
answer = client.qna_search("What is the capital of France?")
print(answer)

Migrate from Tavily

UnSearch is designed as a drop-in replacement for Tavily. Migration is simple:

Option 1: Direct replacement

# Before (Tavily)
from tavily import TavilyClient
client = TavilyClient(api_key="tvly-...")

# After (UnSearch)
from unsearch import UnSearchClient
client = UnSearchClient(api_key="uns-...")

Option 2: Alias (minimal code changes)

# Add this import alias
from unsearch import UnSearchClient as TavilyClient

# Your existing code works unchanged
client = TavilyClient(api_key="uns-...")
response = client.search("query")

LangChain Integration

from unsearch.langchain import UnSearchResults

# Create tool
tool = UnSearchResults(
    api_key="your-api-key",
    max_results=5,
    include_answer=True
)

# Use in agent
results = tool.invoke("What is LangChain?")

Migrate from TavilySearchResults

# Before
from langchain_community.tools import TavilySearchResults
tool = TavilySearchResults(api_key="tvly-...")

# After (option 1)
from unsearch.langchain import UnSearchResults
tool = UnSearchResults(api_key="uns-...")

# After (option 2: alias)
from unsearch.langchain import UnSearchResults as TavilySearchResults
tool = TavilySearchResults(api_key="uns-...")

Self-Hosted Instance

# Point to your self-hosted UnSearch
client = UnSearchClient(
    api_key="your-api-key",
    base_url="https://your-unsearch-instance.com"
)

# Or use environment variable
# export UNSEARCH_BASE_URL=https://your-unsearch-instance.com

Zero-Retention Mode

For privacy-sensitive applications:

# Enable zero-retention (no data stored on server)
client = UnSearchClient(
    api_key="your-api-key",
    zero_retention=True
)

# Or per-request via header
# X-Zero-Retention: true

Async Support

from unsearch import AsyncUnSearchClient

async def search():
    async with AsyncUnSearchClient(api_key="your-key") as client:
        response = await client.search("async search query")
        return response.results

API Reference

Agent API (Tavily-compatible)

Method	Description
search(query, **options)	AI-optimized web search
extract(urls, **options)	Extract content from URLs
research(query, **options)	Multi-step deep research
qna_search(query)	Quick Q&A shortcut
get_search_context(query, max_results)	Get RAG context
list_models()	List available AI models

Core Search API

Method	Description
core_search(query, **options)	Full-featured search
batch_search(queries, **options)	Search multiple queries
list_engines()	List search engines

RAG API

Method	Description
rag_research(topic, **options)	Deep RAG research
rag_search(query, **options)	Quick RAG search
semantic_search(corpus_id, query)	Semantic search over corpus
image_search(query)	Image search
list_corpora()	List research corpora

Neural API (Exa-compatible)

Method	Description
neural_search(query, **options)	Neural/semantic search
find_similar(url=None, text=None)	Find similar content
extract_highlights(query, content)	Extract key highlights
predictive_search(context)	Predict next search

Topic Monitoring API

Method	Description
create_monitor(topic, **options)	Create topic monitor
list_monitors()	List all monitors
get_monitor(monitor_id)	Get a monitor
pause_monitor(monitor_id)	Pause a monitor
resume_monitor(monitor_id)	Resume a monitor
delete_monitor(monitor_id)	Delete a monitor
get_monitor_results(monitor_id)	Get monitor results

Enhanced API (Advanced scraping)

Method	Description
enhanced_search(query, **options)	Advanced search with extraction
enhanced_scrape(urls, **options)	Direct URL scraping
enhanced_features()	List enhanced features
extract_tables(html_content)	Extract tables from HTML
chunk_content(text, strategy)	Chunk text for RAG
discover_urls(base_url, source)	Discover URLs from sitemaps

Knowledge Graph API

Method	Description
knowledge_extract(text)	Extract entities & relationships
knowledge_search(query)	Search knowledge graph
knowledge_people(query)	Search for people
knowledge_get_entity(entity_id)	Get specific entity
knowledge_graph()	Get graph structure

Agent Registration API

Method	Description
register_agent(name)	Register new AI agent (sandbox)
agent_status()	Get agent status
resend_claim()	Resend claim link

Verification API (Fact-checking)

Method	Description
verify_claim(claim, **options)	Verify a claim
check_source_credibility(url)	Check source credibility
batch_verify(claims)	Batch verify claims

Extended Examples

Deep Research

# Multi-step research with AI synthesis
research = client.research(
    query="Impact of AI on healthcare",
    depth="deep",  # quick, standard, deep, comprehensive
    max_sources=20,
    include_analysis=True,
    focus_areas=["diagnostics", "drug discovery"]
)

print(research.executive_summary)
print(research.key_findings)

Neural Search (Exa-compatible)

# Semantic search with auto-prompting
results = client.neural_search(
    query="innovations in renewable energy",
    num_results=10,
    use_autoprompt=True,  # AI expands query
    include_highlights=True,
    category="tech"
)

print("Expanded queries:", results.expanded_queries)
for r in results.results:
    print(r.title, r.highlights)

Topic Monitoring

# Create a real-time monitor
monitor = client.create_monitor(
    topic="artificial intelligence regulations",
    keywords=["AI", "regulation", "EU"],
    check_interval_minutes=60,
    webhook_url="https://your-app.com/webhook",
    deep_analysis=True
)

# Get results
results = client.get_monitor_results(monitor["id"], limit=50)

Fact Verification

# Verify a claim
verification = client.verify_claim(
    claim="The Earth is approximately 4.5 billion years old",
    depth="thorough"
)

print(f"Verdict: {verification.verdict}")
print(f"Confidence: {verification.confidence}%")
print(f"Summary: {verification.summary}")

# Check source credibility
credibility = client.check_source_credibility("https://example-news.com")
print(f"Score: {credibility.credibility_score}")
print(f"Bias: {credibility.bias_rating}")

Why UnSearch over Tavily?

Feature	Tavily	UnSearch
Open Source	❌	✅
Self-Hostable	❌	✅
Zero Retention	❌	✅
Cost at Scale	~$0.0075/query	$0.0003/query
Free Tier	1,000/mo	5,000/mo
Search Engines	Single	70+
Neural Search	❌	✅
Topic Monitoring	❌	✅
Fact Verification	❌	✅

License

Apache 2.0