Skip to main content
​
New include_domains_mode parameter for Search.
  • include_domains_mode controls how include_domains is applied. Set it to "prefer" to also search the rest of the web, so results outside include_domains can still surface, or "restrict" to keep the existing hard-filter behavior. Requires include_domains to be set, or the request returns a 400.
  • Supported domain types: plain domains and subdomains (e.g. wikipedia.org, en.wikipedia.org), with or without a protocol or www. prefix. Wildcards, regex, and URLs with a path, query, or fragment are not supported.
  • Hard-filter matching: under include_domains_mode="restrict" and for exclude_domains, listing a domain (e.g. wikipedia.org) matches that domain and all its subdomains, while listing a subdomain (e.g. en.wikipedia.org) matches only that exact host.
See Best Practices for Search for when to boost trusted domains vs. hard-filter to them.

​
Safe search for everyone

safe_search is now available on all plans. Set "safe_search": true in your Search request to drop adult and unsafe content from results and images. Defaults to false. Not supported for fast or ultra-fast search depths.New language and filter_by_language parameters for Search.
  • language boosts results in a specific language. Accepts an ISO 639-1 code (e.g. en, fr, zh-cn) or an English language name (e.g. english, french).
  • filter_by_language, when true, strictly drops results that don’t match language instead of only boosting them in ranking. Requires language to be set, or the request returns a 400.
See Best Practices for Search for when to use a soft boost vs. a strict filter.

​
Logs API: filter by API key and Research endpoint

The Logs endpoint can now return logs for the calling API key only, and filter by the Research endpoint.
  • filter_by_api_key (boolean, default false): when true, only logs for the API key in the Authorization header are returned, instead of logs for all API keys under your account or organization.
  • research is now a valid value in the endpoints filter, alongside search, extract, map, and crawl.
The Logs API is available on paid plans and is now fully documented in the API reference.
​

​
Chunked content for basic search

search_depth=basic now returns reranked chunks, and chunks_per_source applies to it.
  • Basic search now returns chunks. basic previously returned a single page summary per source. It now returns multiple semantically relevant snippets per URL, so results[].content may contain several chunks joined by [...].
  • chunks_per_source now applies to basic. It previously affected advanced and fast, and now controls the number of chunks per source at those depths plus basic.
  • Choosing between depths. advanced still offers broader coverage and the highest relevance at higher latency. See basic vs advanced.
​

​
Session tracking headers

Attach session and end-user identifiers to any API request so it can be attributed to a session or end-user. See Session Tracking.
  • X-Session-Id: opaque identifier for a session of related requests (e.g. all calls from one user conversation).
  • X-Human-Id: opaque identifier for the end-user behind the request. Useful when a single API key serves many users.
  • Populated automatically by the Tavily MCP server. Local MCP generates a per-process session ID and forwards HUMAN_ID; remote MCP derives the session ID from the MCP initialize handshake.

​
Domain filtering and output length for Research

New include_domains, exclude_domains, and output_length parameters for Research.
  • include_domains (array of string, default [], max 20): a soft preference. The research agent prioritizes these domains, but URLs from other domains can still appear in the report. Matching is host-based and includes subdomains (e.g. sec.gov matches data.sec.gov).
  • exclude_domains (array of string, default [], max 20): a hard blocklist. No URL from a listed domain or its subdomains appears in the response. Matching is downward only: blocking medium.com also blocks blog.medium.com, but blocking blog.medium.com leaves medium.com allowed.
  • output_length (short, standard, or long; default standard): typed control over response size. Ranges are targets, not hard caps.
​

​
Enterprise API key management endpoints

Enterprise users can now manage API keys programmatically:These endpoints are available on the Enterprise plan only. Talk to an expert to learn more.
​

​
Exact match parameter

New exact_match parameter (boolean, default false) for Search. It returns only results containing the exact quoted phrase(s) in your query, bypassing synonyms and semantic variations.
  • Wrap target phrases in quotes within your query, e.g. "John Smith" CEO Acme Corp.
  • Because it narrows retrieval, it may return fewer results, or empty result fields when no exact matches are found.
  • Best suited for due diligence, data enrichment, and legal or compliance use cases that need verbatim matches.
​

​
Project tracking with the X-Project-ID header

Attach a project ID to your API requests to organize and track usage by project. Useful when one API key is used across several projects or applications.
  • HTTP header: add X-Project-ID: your-project-id to any API request.
  • Python SDK: pass project_id="your-project-id" when creating the client, or set the TAVILY_PROJECT environment variable.
  • JavaScript SDK: pass projectId: "your-project-id" when creating the client, or set the TAVILY_PROJECT environment variable.
  • An API key can be associated with multiple projects.
  • Filter requests by project in the /usage endpoint and the platform usage dashboard.
​

​
New search depths: fast and ultra-fast (beta)

The search_depth parameter has two new options:
  • fast (beta): optimized for low latency while keeping high relevance to the query. Costs 1 API credit.
  • ultra-fast (beta): optimized strictly for latency. Costs 1 API credit.

​
Intent-based extraction

New query and chunks_per_source parameters for Extract and Crawl.
  • query (Extract, string): user intent for reranking extracted content chunks by relevance.
  • chunks_per_source (Extract and Crawl, integer, 1 to 5, default 3): the maximum number of relevant chunks returned per source, which controls raw_content length. Chunks are short snippets (up to 500 characters each) and appear in raw_content as <chunk 1> [...] <chunk 2> [...] <chunk 3>. Available only when query (Extract) or instructions (Crawl) is provided.

​
Include usage parameter

Set include_usage (boolean, default false) to true to get credit usage in the response for Search, Extract, Crawl, and Map. The response then includes a usage object with credits information.The value may be 0 if total successful calls haven’t reached the minimum threshold. See Credits & Pricing for details.
​

​
Vercel AI SDK v5 integration

The new @tavily/ai-sdk package provides ready-to-use tools for Vercel’s AI SDK v5: tavilySearch, tavilyExtract, tavilyCrawl, and tavilyMap. It has full TypeScript support. See the integration guide to get started.

​
Crawl and Map timeout parameter

New timeout parameter for Crawl and Map to control how long to wait before the operation times out.
  • Type: float
  • Range: 10 to 150 seconds
  • Default: 150 seconds
​

​
Team roles and permissions

You can now assign roles to team members. Each team has one owner, and can have multiple admins and members. Roles differ in their Billing and Settings permissions:

​
Extract timeout parameter

New timeout parameter for Extract to control how long to wait for URL extraction before timing out.
  • Type: number (float)
  • Range: 1.0 to 60.0 seconds
  • Default: if not set, the timeout depends on extract_depth: 10 seconds for basic and 30 seconds for advanced.
​

​
Start date and end date parameters

New start_date and end_date parameters for Search, both in YYYY-MM-DD format.
  • start_date returns results after the given date.
  • end_date returns results before the given date.
  • For example, set start_date to 2025-01-01 and end_date to 2025-04-01 to get results strictly from that range.

​
Usage dashboard

Paid users and teams can log in to view the usage dashboard:
  • Usage graph: daily usage across all Tavily endpoints, with history for month-over-month usage and spend comparison.
  • Logs table: per-request detail for visibility and traceability of every Tavily call.
​

​
Include favicon parameter

Set include_favicon to true to get the favicon URL (if available) for each result in Search, Extract, and Crawl. Useful for showing site icons next to results.

​
Auto parameters

New auto_parameters parameter (boolean, default false) for Search. When enabled, Tavily configures search parameters automatically based on the query’s content and intent. Parameters you set explicitly override the automatic ones.
  • include_answer, include_raw_content, and max_results must always be set manually, since they affect response size.
  • search_depth may be set to advanced automatically when that’s likely to improve results, which costs 2 API credits per request. Set search_depth to basic explicitly to avoid the extra cost.
​

​
Usage endpoint

Check your API usage and plan limits with the /usage endpoint: send GET https://api.tavily.com/usage with your API key to monitor your account in real time.

​
Country parameter

New country parameter for Search boosts results from a specific country. Available only when topic is general.

​
Make and n8n integrations

  • n8n: add real-time web search and content extraction to your n8n workflows without writing code.
  • Make: bring real-time online information into your Make workflows and automations without writing code.

​
Markdown format for Extract

New format parameter for Extract sets the format of extracted content: markdown (default) or text. text returns plain text and may increase latency.
​

​
Advanced search and chunks per source

New search_depth and chunks_per_source parameters for Search.
  • search_depth (basic or advanced, default basic): advanced retrieves the most relevant sources and content snippets for your query, while basic returns generic snippets from each source. basic costs 1 API credit; advanced costs 2.
  • chunks_per_source: the maximum number of relevant chunks (short snippets up to 500 characters each) returned per source, which controls content length. Chunks appear in the content field as <chunk 1> [...] <chunk 2> [...] <chunk 3>. Available only when search_depth is advanced. Range: 1 to 3.

​
Tavily Crawl (beta)

Tavily Crawl traverses a website like a graph: it starts from a base URL and discovers and extracts content from linked pages. You can:
  • Set the starting URL and let the crawler follow links to map the site structure.
  • Control the depth and breadth of the crawl to focus on specific sections or cover the whole site.
  • Apply filters and custom instructions to target the most relevant pages or content types.
  • Aggregate extracted content for analysis, reporting, or your own workflows.
Crawl suits large-scale content aggregation, competitive research, and knowledge base creation. See the Crawl API reference for details and examples.