What’s New

We’ve added significant improvements to company search due to a fine-tuned retrieval model and entity-matching pipeline. Use type = "auto" , category = "company" to use this in our search API.

Date: January 21, 2026

We’ve added significant improvements to company search due to a fine-tuned retrieval model and entity-matching pipeline for this vertical of queries.
Try Company Search in our API Playground with type = "auto" , category = "company" .

State-of-the-art company search: We fine-tuned our retrieval model specifically for company search and built an ingestion pipeline optimized for entity matching—delivering accurate results across attributes like industry, geography, funding stage, and employee count.
Use case focused: Run queries like “fintech companies in Switzerland” or “Japanese AI companies founded in 2023” and programmatically enrich results with structured company data for sales prospecting, market research, and supply chain workflows.

How to Use Company Search

Use type="auto" and category="company" in your search requests:

curl -X POST https://api.exa.ai/search \
-H "x-api-key: EXA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "Agtech companies optimizing pesticide placement with computer vision",
"type": "auto",
"category": "company",
"numResults": 10
}'

The company category supports queries across:

• Named lookup: “Sakana AI company” or “Tell me about exa.ai”
• Attribute filtering: Industry, geography, founding year, employee count
• Funding queries: Stage, amount raised, recent rounds
• Composite queries: Multiple constraints like “Israeli security companies founded after 2015”
• Semantic queries: Natural language descriptions like “Companies like Bell Labs”

Structured Entity Data

With this launch, we are introducing entities for company search, a new primitive from Exa that will return high quality, structured information from web data.
Company search results now include structured entity data with detailed company information:
{
"entities": [
{
"id": "https://exa.ai/library/company/metaphor-systems",
"type": "company",
"version": 1,
"properties": {
"name": "Exa",
"foundedYear": null,
"description": "Exa was built with a simple goal — to organize all knowledge...",
"workforce": {
"total": 48
},
"headquarters": {
"address": "430 Shotwell St",
"city": "San Francisco",
"postalCode": "94110",
"country": "United States"
},
"financials": {
"revenueAnnual": null,
"fundingTotal": 107000000,
"fundingLatestRound": {
"name": "Series B",
"date": "2025-09-03",
"amount": 85000000
}
},
"webTraffic": {
"total": 477156
}
}
}
]
}

### Learn More

- Read our blog post: Introducing Exa’s Company Search Benchmark
- Try it in the API Playground

### Need Help?

If you have questions about company search or want to learn more about optimizing your queries, reach out to [email protected]. We’re here to help you get the most out of Exa Company Search!

Original source Report a problem

We’re launching state-of-the-art people search with 1B+ indexed profiles. The ‘linkedin’ category is now replaced with ‘people’ for better results.

Date: December 19, 2025

We’re launching Exa People Search, a new way to find and discover people on the web, designed for real production use across sales, recruiting, research, and more.

Try People Search in our API Playground with category = "people".

What’s New

State-of-the-art people search: Exa indexed 1B+ public profiles and trained a hybrid retrieval system (fine-tuned embeddings + Exa Search) to deliver highly accurate role, skill, and company based people search at web scale.

Usecase focused: Customers can run queries like “VP of Product at Microsoft” or “enterprise sales reps from Microsoft in EMEA” and programmatically enrich results with profiles for sales, recruiting, and market research workflows.

What Changed

We’re replacing the linkedin category with the new people category to provide better, more comprehensive people search results.

Before

Old approach - limited to LinkedIn

result = exa.search("VP of Product at Microsoft", category = "linkedin")

After

New approach - comprehensive people search across the web

result = exa.search("VP of Product at Microsoft", category = "people")

How to Use People Search

Simply use category="people" in your search requests:

curl -X POST https://api.exa.ai/search \
-H "x-api-key: EXA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
  "query": "Product managers at Microsoft",
  "category": "people",
  "numResults": 10
}'

The new people category provides:

• Broader coverage: 1B+ profiles across the entire web, not just LinkedIn
• Better accuracy: Fine-tuned embeddings specifically for people search
• More relevant results: Hybrid retrieval system optimized for role, skill, and company queries

Learn More

• Read our blog post: Introducing Exa’s People Search Benchmarks
• Follow our announcement: Twitter/X
• Try it in the API Playground

Need Help?

If you have questions about migrating from the linkedin category or want to learn more about optimizing your people search queries, reach out to [email protected]. We’re here to help you get the most out of Exa People Search!

Original source Report a problem

The highlights feature has been reintroduced in the JavaScript SDK (exa-js) as of version 2.0.11.

Date: November 26, 2025

The highlights feature is back in the JavaScript SDK. Following user feedback, we’ve reintroduced highlights in exa-js v2.0.11, allowing you to extract key sentences from search results with relevance scores.

What’s Back

The highlights option is now available in search and contents operations:

• highlights: true - Returns highlighted sentences with default settings
• highlights: { maxCharacters, query } - Customize extraction behavior

Results include:

• highlights: string[] - Array of extracted key sentences
• highlightScores: number[] - Relevance scores for each highlight

Usage Examples

Basic highlights

const results = await exa.searchAndContents("latest AI research", {
  highlights: true
});
console.log(results.results[0].highlights);
// ["Key sentence from the article...", "Another relevant excerpt..."]

With options

const results = await exa.searchAndContents("machine learning tutorials", {
  highlights: {
    maxCharacters: 2000,
    query: "beginner friendly"
  }
});

Combined with text

const results = await exa.searchAndContents("climate news", {
  text: true,
  highlights: true
});
// Returns both full text and highlighted excerpts

Scope

This update applies only to the JavaScript SDK (exa-js). Other SDKs can access highlights via direct API calls.

Installation

npm install exa-js@latest

Original source Report a problem

Introducing Exa Deep: Get better results with smart query expansion and high-quality summaries.

Date: November 20, 2025

We’re excited to introduce Exa Deep - a new search type that finds better results by running multiple searches at once and gives you high-quality context for each result. You can send just one query (we’ll create variations automatically) or provide your own query variations using the additionalQueries parameter for even better results.

Deep search is available on our API Playground.
Try Deep search in the dashboard →

How Deep Search Works

When you use Deep search, here’s what happens:

• Query Expansion: If you only send one query, we automatically create variations. If you send query variations yourself using additionalQueries, we use those instead. For best results, consider having a good LLM model (like GPT-5 or Claude 4.5 Sonnet) generate the query variations for you.
• Parallel Search: We search for your main query and all variations at the same time
• Smart Ranking: We combine and rank all results to give you the most relevant ones
• Summary Generation: Each result gets a detailed, accurate summary

How to Use Deep Search

Using Deep search is simple - just add type="deep" to your search requests. You can also add additionalQueries for even better results:

Basic Deep Search:

curl -X POST https://api.exa.ai/search \
-H "x-api-key: EXA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
  "query": "blog post about AI",
  "type": "deep",
  "contents": {
    "text": true,
    "context": true
  }
}'

Deep Search with Query Variations:

curl -X POST https://api.exa.ai/search \
-H "x-api-key: EXA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
  "query": "blog post about AI",
  "additionalQueries": ["AI blogpost", "machine learning blogs"],
  "type": "deep",
  "contents": {
    "text": true,
    "context": true
  }
}'

What You Get Back

The context parameter shown in examples below is now deprecated. Use highlights or text instead.

Deep search returns a context field that gives you detailed context:
{
"requestId": "975a6ff95c69a0bdc558f01c99ede801",
"context": "Title: AI News | Latest AI News, Analysis & Events\nPublished Date: 2025-11-19T10:22:05.000Z\nURL: https://www.artificialintelligence-news.com/\nSummary: This page is a collection of recent news, analysis, and resources related to Artificial Intelligence (AI). It features various blog posts and articles covering topics such as:\n\n* Workforce & HR: Using ChatGPT for team planning and the Royal Navy's use of AI in recruitment.\n* Technology & Infrastructure: The role of Pure Storage and Azure in AI-ready data, and alliances between Microsoft, NVIDIA, and Anthropic for AI compute.\n* AI Models & Tools: Google’s Veo 3 video creation tools, Samsung’s small AI model, and lightweight LLMs for enterprise use in Japan.\n* Industry Applications: AI in asset management (Franklin Templeton & Wand AI), Levi Strauss's DTC model, and accounting firms using AI agents.\n* Research & Hardware: Breakthroughs in analog AI chips for deep learning a....",
"results": [
{
"id": "https://www.artificial-intelligence.blog/",
"title": "AI Blog - Artificial Intelligence Blog",
"url": "https://www.artificial-intelligence.blog/",
"publishedDate": "2016-01-01T00:00:00.000Z",
"author": "Artificial Superintelligence (ASI)",
"text": "This is a blog dedicated to Artificial Intelligence, aiming to keep readers updated on the field in innovative, informative, and entertaining ways. The content, including images and text, is largely generated by AI. The blog features various posts covering topics such as new AI models, societal impact, technical discussions, industry focus, market analysis, and beginner guides...",
"image": "http://static1.squarespace.com/static/62ec2bc76a27db7b37a2b32f/t/68974039b71d0f1a6f3e6920/1754742841664/of+ai-blog-youtube-2025.png?format=1500w",
"favicon": "https://images.squarespace-cdn.com/content/v1/62ec2bc76a27db7b37a2b32f/abfb4587-35b3-411f-8603-7e24344b95fc/favicon.ico?format=100w"
},
{
"id": "https://blog.google/technology/ai/",
"title": "AI",
"url": "https://blog.google/technology/ai/",
"publishedDate": "2018-06-07T18:13:42.000Z",
"author": "",
"text": "The work we're doing to make AI helpful for everyone. This page covers Gemini Models, Gemini App updates, Research developments, and Developer resources. Recent highlights include the release of Gemini 3, tips for using Nano Banana Pro, and various AI research initiatives...",
"image": "https://blog.google/static/blogv2/images/google-200x200.png?version=pr20251113-1736",
"favicon": "https://blog.google/favicon.ico"
}
],
"searchTime": 2619.8,
"costDollars": {
"total": 0.022,
"search": {
"neural": 0.005
},
"contents": {
"text": 0.017
}
}
}

### Available Search Types

Now you have four search types to choose from:
- Auto (default): Our best search, intelligently combines multiple search methods
- Fast: Fastest search with lowest latency
- Deep: Deep search with query expansion and summaries
- Neural: Predicts the most relevant results based on query meaning

We’re excited for you to try Deep search and see how it can improve your search results!

Original source Report a problem

Language filtering is now turned on for everyone by default

Exa now detects your query language and only searches web search results in the same language.

Date: November 5, 2025

We now return results that match the language of your search query. This feature is now turned on by default for all users.

What Changed

Before this update, Exa would mostly return results in your query language, but you might also get results from other languages mixed in. Now, Exa detects what language you’re searching in and only shows you web search results in that same language.

What This Means for You

1. Better results: Your search results will now be more relevant because they match your query language.
2. No setup needed: This feature is already working for your account - you don’t need to do anything.
3. Works across languages: Whether you search in English, Spanish, French, or any other language, you’ll get results in that language.

How It Works

When you search, Exa automatically:

• Detects what language your query is in
• Searches only through web search results in that same language
• Returns results that are more relevant to your language needs

This update makes search results much more useful, especially if you work with content in multiple languages. The filtering works particularly well and has shown big improvements in our testing.

If you have any questions about this update, please reach out to [email protected].

Original source Report a problem

Major SDK update

Major SDK update with contents included by default in search, highlights feature removed from SDKs, and use_autoprompt field deprecated in all API responses.

We’re releasing a major version update to our SDKs along with changes to API responses. This update makes content retrieval more convenient while removing deprecated features.

1. Contents Included by Default in SDKs

Search operations in the SDKs now include page contents by default, eliminating the need for a separate contents call in most workflows. You can opt out if you need faster searches without content.

2. Highlights Feature Removed from SDKs

The highlights feature has been completely removed from all SDKs. This feature was previously deprecated and is no longer available in the SDK packages.

Update (November 2025): Highlights have been reintroduced in the JavaScript SDK as of exa-js v2.0.11. See JS SDK: highlights restored for details.

Migration Options:

• Option 1: Do not upgrade to the new major version if you still need highlights
• Option 2: Use the API directly to access highlights functionality
• Option 3: Use “AI Summary” to get the main summary of “text”

3. use_autoprompt Deprecated in All API Responses

The use_autoprompt field has been deprecated and removed from all API responses across the entire platform. This field is no longer needed with current search improvements.

Need Help?

If you have questions about upgrading or need help with migration, please reach out to [email protected]. We’re here to help ensure a smooth transition to the new major version.

Original source Report a problem

Date: August 4, 2025

What’s New

The includeDomains and excludeDomains parameters now support:

• Path-specific filtering: Target specific sections of a domain by including the path
• Subdomain wildcard matching: Use *.domain.com to match all subdomains

Examples

Pattern | What it matches | Example URLs
"*.substack.com" | Any subdomain of substack.com | https://thehobbyist.substack.com/p/location-matters-6-days-273-bets
"exa.ai/blog" | Only the blog section of exa.ai | https://exa.ai/blog/meet-the-exacluster
"linkedin.com/company" | Company profiles on LinkedIn | https://www.linkedin.com/company/exa-ai

When to Use Path Filtering

Path filtering is useful for things like:

• Blogs: Search within blogs like stripe.com/blog, openai.com/blog, or stratechery.com/2025
• Product Catalogs: Query product pages like amazon.com/dp, etsy.com/listing, or ikea.com/us/en/cat
• Directories: Search specific directories like ycombinator.com/companies, crunchbase.com/organization, or github.com/orgs

How To Use Path Filtering

You can use the same includeDomains and excludeDomains parameters:

Python example code snippet provided in the original content.

Need Help?

If you have any questions about domain filtering or need help with your specific use case, please reach out to [email protected].

Original source Report a problem

userLocation added to the search API to bias search results based on geographic location.

Date: July 30, 2025

We’re excited to announce a new userLocation parameter that lets you bias search results based on a user’s geographic region. The location is passed as an ISO 3166-1 alpha-2 country code (e.g., “fr” for France, “us” for the United States). If this field is provided, search will return results that are more relevant to users in the provided region.

When to Use Geolocation Filter
The userLocation parameter is particularly useful for:

• Multi-regional applications: Show users content that’s relevant to their region
• Language-specific content: Prioritizing content in regional languages
• Local discovery: Surface products or businesses relevant to the users region

Consider using geolocation filtering when the user’s physical location or regional context significantly impacts the relevance of search results.

How To Use Geolocation Filter
Here’s how to implement the new userLocation parameter:

Python example:

result = exa.search_and_contents(
    "football rules",
    type = "auto",
    livecrawl = "never",
    userLocation = "us",  # ISO 3166-1 alpha-2 country code
    num_results = 10
)

Response Structure Changes
The response structure remains unchanged - geolocation filtering affects result ranking and relevance scoring, but doesn’t modify the response format.

Need Help?
If you have any questions about location filtering or need help with your specific use case, please reach out to [email protected].

Original source Report a problem

Introducing Exa Fast: The world’s fastest search API.

Date: July 29, 2025

We’re excited to introduce Exa Fast - the fastest search API in the world. Exa Fast uses streamlined versions of our search models with p50 latency below 425ms.

Fast search is available immediately on all API plans.
Try Fast search in the dashboard →

What’s New

The Fast search type provides:

• Speed: p50 latency below 425ms - that’s 30% faster than other search APIs
• Exa Index: Uses the same index of high quality content as our neural search
• Customization: Full compatibility with all the same parameters as our other search types

When to Use Fast Search

Fast search is ideal for:

• Fast web grounding: Integrate real-time web information into responses without sacrificing speed and impacting user experience
• Agentic workflows: AI agents like deep research that use dozens or hundreds of search calls where milliseconds add up
• Low-latency AI products: Latency-sensitive applications like AI voice companions where every millisecond matters

How to Use Fast Search

Using Fast search is simple - just add type="fast" to your search requests:

result = exa.search_and_contents(
    "latest AI news",
    type = "fast",
    livecrawl = "never",
)

Options That Impact Latency

While Fast search is optimized for speed, certain options can increase response times:

• Live crawling: Fetching content live requires real-time web requests. Set livecrawl="never" to use cached content and maintain optimal speed.
• AI summaries: Requesting AI-generated summaries requires LLM processing, which adds significant latency to your requests.
• Complex date filters: Using wide date ranges or multiple date constraints requires additional filtering that can slow down results.
• Include/exclude text: Text-based content filtering requires scanning through results, which impacts response times.
• Subpages: Including subpages in your search requires additional processing and can significantly increase latency.

For the fastest possible performance, use Fast search with minimal parameters and rely on cached content.

Original source Report a problem

Date: July 21, 2025

We’re deprecating relevance scores in Auto search due to architectural improvements. Scores will remain available in Neural search.

What Changed

Previously, Auto and Neural search types returned relevance scores - a number from 0 to 1 representing similarity between the query and each result. With our new Auto search architecture, we can no longer generate meaningful scores for Auto search results.
The search functionality works exactly the same way as it did before - you’ll still get the same high-quality results, just without the score field in the response.

What This Means for You

• Auto search: The score field will no longer be returned in search results
• Neural search: Scores continue to work exactly as before with no changes
• Migration needed: If your application relies on scores from Auto search, you should migrate as soon as possible

How to Update Your Code

If you currently use scores from Auto search, here is what you can do:

Remove Score Dependencies

# Before: Code that depends on scores
result = exa.search("AI startups", type = "auto")
sorted_results = sorted(result.results, key = lambda x : x.score, reverse = True)

# After: Use results in the order returned (already optimally ranked)
result = exa.search("AI startups", type = "auto")
# Results are already ranked by relevance, no need to sort by score
for item in result.results:
    print (f"Title: {item.title}")

Response Structure Changes

Auto Search (New)

{
  "results" : [
    {
      "title" : "Example AI Startup",
      "url" : "https://example-startup.com",
      "id" : "abc123",
      "publishedDate" : "2024-01-15",
      "author" : "John Doe"
      // Note: No 'score' field
    }
  ]
}

Neural Search (Unchanged)

{
  "results" : [
    {
      "title" : "Example AI Startup",
      "url" : "https://example-startup.com",
      "id" : "abc123",
      "publishedDate" : "2024-01-15",
      "author" : "John Doe"
    }
  ]
}

Need Help with Migration?

If you have questions about migrating from Auto search scores or need help determining the best search type for your use case, please reach out to [email protected]. We’re here to help ensure a smooth transition.

Original source Report a problem