Last updated: 2026-10-03
Tavily is a search API for AI agents that returns reranked web results, page extraction, crawling and research through one credit balance. Nebius announced an agreement to acquire it in February 2026 and names IBM, Cohere and Groq among its customers. The vendor docs list a built-in Hermes Agent integration.
About Tavily
Tavily is a web access API built for AI agents. An agent sends a question or a URL and gets back reranked snippets, cleaned page content or a finished research report, so the model reasons over fresh context instead of raw HTML. Founder and CEO Rotem Weiss runs it, and Nebius announced an agreement to acquire the company on 2026-02-10. The site now carries the "Tavily by Nebius" mark and the brand continues under its own name.
The product is five endpoints on one credit balance. Search returns ranked results with a depth setting, Extract pulls clean content from a list of URLs, Map discovers the URLs on a site, Crawl combines mapping and extraction, and Research runs a longer multi-step task on a mini or pro model. Nebius reported more than 3 million monthly SDK downloads and a developer community above one million users in February 2026, with IBM, Cohere and Groq named as customers.
It is most useful as the web layer under an agent. Tavily's own docs list a built-in integration for Hermes Agent with keyless access, and the same API suits OpenClaw builds. A remote MCP server connects it to Claude Code and Cursor, and the SDKs plug into LangChain, LlamaIndex and n8n workflows. If your agent starts from a question, compare Exa and Perplexity. If it starts from a URL and needs page interaction, look at Firecrawl or You.com.
Pricing is credit based. A basic search costs 1 credit and an advanced search costs 2, while extraction is billed per five successful URLs. The Free plan gives 1,000 credits a month with no card, and paid plans run from Project to Growth before Enterprise. Credits reset on the first of each month, and requests stop when they run out unless pay as you go is on. The API is closed source, though the SDKs are public on GitHub, so you cannot self-host the search itself.
Pricing
Free (Researcher) plan with 1,000 credits a month, no card. 008 per credit. Monthly plans: Project $30 (4,000 credits), Bootstrap $100 (15,000), Startup $220 (38,000), Growth $500 (100,000).
Enterprise is custom with custom rate limits and SLAs. Students use it free. A basic search is 1 credit, an advanced search 2, extraction 1 credit per 5 URLs on basic depth.
Credits reset on the 1st of each month. Production keys need a paid plan or pay as you go. Prices in USD, read from the vendor pricing and credits pages on 2026-10-03.
| Tier | Monthly price | What it includes |
|---|---|---|
| Researcher (Free) | Free | 1,000 API credits a month, no credit card, email support |
| Project | $30/mo | 4,000 credits a month ($0.0075 per credit), higher rate limits, email support |
| Bootstrap | $100/mo | 15,000 credits a month ($0.0067 per credit) |
| Startup | $220/mo | 38,000 credits a month ($0.0058 per credit) |
| Growth | $500/mo | 100,000 credits a month ($0.005 per credit) |
| Pay as you go | Custom | $0.008 per credit, cancel anytime, email support |
| Enterprise | Custom | Custom API calls, custom rate limits, enterprise support and SLAs, enterprise security and privacy |
Key Features
- Search with depth control: One call returns reranked, distilled snippets, billed at 1 credit for basic depth and 2 credits for advanced.
- Extract: Returns cleaned page text for a batch of URLs, billed 1 credit per 5 successful extractions on basic depth, with failed fetches free.
- Map and Crawl: Map discovers the URLs on a site and Crawl adds content extraction, billed as the sum of the two.
- Research endpoint: Runs a multi-step research task on a mini or pro model, billed between 4 and 250 credits per request depending on the model.
- Remote MCP server: A hosted server at mcp.tavily.com connects Claude, Cursor and other MCP clients without a local install.
- Keyless access: Search and Extract work for free with no account or key, at a lower rate limit, until you sign up for a free key.
Pros
- You can prototype on the free plan with no card, and credits reset on the first of every month.
- Five endpoints share one credit balance, so search, extraction, mapping, crawling and research need no second vendor account.
- Production keys allow ten times the request rate of development keys, which suits agents that burst.
- Listed in the AWS AgentCore tool catalog, the Azure MCP Center and the Snowflake and Databricks marketplaces, per the vendor product page.
Cons
- The search API is closed source, so you cannot self-host it, unlike an open source core such as Firecrawl.
- Production-rate keys require a paid plan or pay as you go, so a free-tier prototype stays on the lower development limit.
- Research requests are billed dynamically, so a single pro-model task can use a large share of the free monthly allowance.
- Nebius announced the acquisition in February 2026, so roadmap and pricing may shift under new ownership.
Data Handling
- Compliance
- SOC 2 (badge on the vendor pricing page · linked to its trust center)
Frequently Asked Questions
How much does Tavily cost in 2026?
Tavily bills in credits. The monthly plans are Project at $30 for 4,000 credits, Bootstrap at $100 for 15,000, Startup at $220 for 38,000 and Growth at $500 for 100,000, while pay as you go is $0.008 per credit once a plan runs out. Enterprise pricing is custom. A basic search uses 1 credit and an advanced search uses 2, so the real cost depends on search depth and how much you extract.
Does Tavily have a free tier, and what are the limits?
Yes. The Researcher plan includes a monthly credit allowance with no card, resetting on the first day of each month. When the credits run out, requests stop until the reset or an upgrade. The higher production rate limit is unlocked by upgrading, and the docs also describe a keyless mode that is free and rate limited.
What are the best Tavily alternatives for agent search?
Exa suits agents that start from a question and want ranked retrieval with its own index. Perplexity fits when you want a cited answer rather than raw results. Firecrawl is the pick when your agent starts from URLs and needs scraping, crawling or page interaction.
Tavily or Firecrawl: which fits an agent better?
Tavily is the shorter path when the agent asks a question and wants reranked snippets back in one call. Firecrawl is stronger when the agent already has URLs, needs to click through pages, or you want an open source core you can host yourself. Many agent stacks use one for search and the other for deep page extraction.
How do you connect Tavily to an agent like Hermes?
Create a free account, copy the API key from the dashboard, and either install the Python or JavaScript SDK or point your client at the remote MCP server. Tavily documents a built-in integration for Hermes Agent that works with keyless access, so a first search needs no signup. Add the key when you hit the keyless cap.
Top Alternatives
- Exa: Pick Tavily for reranked snippets and extraction on one credit balance; pick Exa when you want ranked retrieval from its own index.
- Firecrawl: Pick Tavily when the agent asks a question; pick Firecrawl when it starts from URLs and needs scraping or page interaction.
- Perplexity: Choose Tavily when you want raw context for your own model; choose Perplexity when you want a finished cited answer.