Skip to the page

One platform to build, search, extract, enrich, map, and monitor data across the entire web.

Get started free

State of the art on the live web

The most capable engine for deep search on the live web, setting a new bar in research, extraction, and search. On WideSearch it outscores every search API at under a tenth of the cost.

Research
SpecStreamParallelExaPerplexityTavilyLinkup
Wide researchWideSearch64.0all 200 tasks57.656.154.055.9—
Wide researchWideSearch, low-cost agent60.945.553.047.047.9—
Conflicting sourcesSeal-066.756.8Ultra 8x45.0Research Pro38Deep Research24Research61Research
Factual accuracySimpleQA Verified9997919592—
Factual accuracySimpleQA Verified, low-cost agent9694919494—

Other providers' figures are their best published scores. WideSearch is item F1; ours on all 200 official tasks with the benchmark's scorer. Low-cost agent rows are scores on GPT-5.6 Luna, on a sample of 100 tasks or questions. Seal-0 figures are each provider's research product, as Parallel and Linkup published them. A dash means no published score.

WideSearchScore vs. cost
4045505560650Item F11020501002005001,000Cost per 1,000 tasks (USD, log scale)SpecStream, all 200 tasks: 64.04045505560650Item F110501005001,000Cost per 1,000 tasks (USD, log scale)SpecStream, all 200 tasks: 64.0

WideSearch gives an agent broad research tasks and checks every cell of the table it returns. On the low-cost agent, at $39 per 1,000 tasks, our engine scores 60.9, above every other API on the frontier agent at $547 to $1,072. On all 200 official tasks it scores 64.0, the highest score any search API has published.

Seal-0Accuracy by provider
  • SpecStream
  • Other providers
010203040506070Accuracy (%)Tavily(Research)24Perplexity(Deep Research)38Exa(Research Pro)45.0Parallel(Ultra 8x)56.8Linkup(Research)61SpecStream66.7010203040506070Accuracy (%)Tavily24Perplexity38Exa45.0Parallel56.8Linkup61SpecStream66.7

Seal-0 asks 111 questions where search results conflict or mislead. Our engine answers 66.7% correctly, above the research product of every provider shown.

SimpleQA VerifiedAccuracy by provider
  • SpecStream
  • Other providers
808590951000Accuracy (%)Exa91Tavily92Perplexity95Parallel97SpecStream99808590951000Accuracy (%)Exa91Tavily92Perplexity95Parallel97SpecStream99

SimpleQA Verified asks short factual questions with one correct answer. Our engine scores 99.

Reading pages
SpecStreamZyteDiffbotrs-trafilaturaTrafilaturaReadability
Article extractionScrapinghub benchmark0.9710.970AutoExtract0.9510.9700.9580.947
Page extractionWCXB0.914——0.9030.8410.726

Other extractors' figures are the scores published by each benchmark. Both benchmarks score F1 against hand-labelled page text. A dash means no published score.

Scrapinghub article benchmarkF1 by extractor
  • SpecStream
  • Other extractors
0.900.920.940.960.980F1Readability0.947Diffbot0.951Trafilatura0.958rs-trafilatura0.970Zyte(AutoExtract)0.970SpecStream0.9710.900.920.940.960.980F1Readability0.947Diffbot0.951Trafilatura0.958rs-trafilatura0.970Zyte0.970SpecStream0.971

The Scrapinghub benchmark scores how exactly an extractor returns the text of an article. Our Extract API scores 0.971, above the best commercial and open-source extractors, and is first on WCXB, the 511-page extraction benchmark, at 0.914.

Scraping for agents
SpecStreamContext.devSpiderScrapflyStringLinkupFirecrawl
Scraping for agentsscrapingbench, tasks done40/4038/4037/4036/4035/4034/4032/40
Scraping for agentsscrapingbench, answer in the response40/4038/4038/4036/4038/4037/4036/40

Other providers' figures are the board's published scores. scrapingbench is its scrape track: 40 tasks on Claude Haiku 5.5 with the board's prompt and scorer. Our run is our own, with each answer checked against the official source on the day of the run.

scrapingbenchTasks done by provider
  • SpecStream
  • Other providers
283032343638400Tasks done (of 40)Firecrawl32Linkup34String35Scrapfly36Spider37Context.dev38SpecStream40283032343638400Tasks done (of 40)Firecrawl32Linkup34String35Scrapfly36Spider37Context.dev38SpecStream40

scrapingbench gives one agent a single tool, the provider's scraper, and 40 tasks on live documentation sites. A script checks every answer against the official source. With our scraper the agent finishes all 40 tasks. No other provider on the board finishes more than 38.

Describe the data you need. Get it live from the web.

Ask for any dataset in plain words, and SpecStream builds it from the entire live web: every match found, extracted and enriched into structured records, then monitored for change.

Turning the entire web into a live database

Search, read, extract, research, and browse the web in real time, from the workspace or from your own code.

  1. See the Search API

Give your agents the most accurate web search API in the world

Your agents ask in plain words and get back finished, structured data, not pages to read.

In every agent and AI chat

Add one URL to Claude, ChatGPT, Cursor or any MCP client. Or build SpecStream into your own agents with the API, A2A and our SDKs.

Agents that pay on their own

Your agent sees the price first and pays per row, by card or in USDC, with no account. On your account, it stays inside the limits you set.

Work that outlasts the session

A run keeps going until the list is complete, for up to four hours.

Connect your agent in one command

Pick your client and copy the line. Read the docs. Are you an AI agent? Start at specstream.ai/agents.md.

Then run /mcp to sign in.
claude mcp add --transport http specstream https://api.specstream.ai/mcp

Ask in Slack. Land it in your CRM.

Ask where your team works

Message SpecStream in Slack or Microsoft Teams. Approve the price in the chat, and the data comes back right there.

Land it in your stack

The data flows into your CRM, sheets or warehouse, and new and changed records keep arriving. In your CRM, records match the ones you already have, so nothing lands twice.

Every integration is included on every plan. See all integrations

Put the whole web to work for every team

Ask in plain words. Get every match on the live web, in structured rows.

Sales and GTM

Your best accounts share things no filter menu lists: the tools they run, the roles they're hiring for, the new markets they've entered. Describe them the way you'd brief a new rep. Get every account that fits, its buying committee and how to reach each person.

Investors and M&A

The best deals are the ones no database lists yet: founder-owned, unfunded, off-market. Write your thesis in a sentence, get every company that fits it, and hear the day one of them moves.

Analysts and newsrooms

The dataset your story or your model needs usually doesn't exist until someone builds it. Ask for it in a sentence. Get the table, with every number traced to its primary source.

Commerce and brands

Your whole category on one screen: every product, price, promotion and stock level across the stores that carry it. Spot MAP breaks, new sellers and empty shelves the day they happen.

Procurement

Second sources, new regions, gaps in your approved vendor list. Describe the part, the standard and the region once. Get every qualified maker, with the certifications their own sites state.

Research

A landscape review has to cover the papers, reports, filings and grey literature no single index holds. Turn your question into one table of everything published on it, each finding linked to its source.

Risk and compliance

KYB, vendor due diligence and adverse media for counterparties, vendors and new clients. One name to everything public about them, from their own site and the press to the official register.

Recruiting

Talent maps for the roles no Boolean string can fill. Find people by what they've built, written and shipped, in code, papers, talks and patents, not by the keywords on their profile.

Product and AI

Enrich every signup and fill every field your product needs. Give your product and your agents any data the web holds, live, through one API.

Government contracts

Federal databases cover state and local portals poorly. Describe your niche once, and its open tenders reach your pipeline with agency, due date and the link to the tender, rechecked on your schedule.

Nonprofit fundraising

A new program needs funders. Name the program and region, and get every funder of that cause there, with focus, typical gift, deadline and contact, each from its 990 or site.

Creator marketing

When search returns creators every rival found, name the niche, market and language, and get creators posting about it in your sheet, with the business contact they published.

Security for your team. Control for your agents.

Agents only work in the enterprise when you decide what they can do, what they can spend and for how long.

  1. Spend your CFO can predict

    Every run shows its price first and never goes past the maximum you approve. Set limits for each member, project and key.

  2. Agents your engineers control

    Give an agent only the operations you choose. Cap it per run and per month, and set the day its access ends. Pages SpecStream reads reach our models as data, never as instructions.

  3. Access your security team can audit

    SAML single sign-on, SCIM provisioning and a 12-month audit log of every sign-in, access change, spend and export.

  4. Terms your legal team can sign

    Hosted in the EU and encrypted in transit and at rest, under a GDPR data processing agreement. No model is trained on your work.

Read our security overviewData processing agreement

5¢

per finished row

Up to 5 facts a row.

$5 per 1,000 searches, $1 on the fast tier. $0.80 per 1,000 pages. A research report from $0.10. Rows billed when delivered. No seats or setup fees.

$10 freeof data on sign-up. No card required.
5¢ an emailand 15¢ a phone number, on any row.
Half the ratefor a refreshed row, and 1¢ for an unchanged one.
From $19 a monthfor plans. Custom pricing for large volumes.

Questions

What is SpecStream?

SpecStream is the live web data engine for teams and agents. Ask for any dataset in plain words, and SpecStream builds it from the entire live web: every match found, extracted and enriched into structured records, then monitored for change. It runs through the API and MCP, and inside Sheets, CRMs, Slack, n8n, Zapier and the other tools teams already use.

Who uses SpecStream?

Sales and deal teams, investors, analysts and newsrooms, commerce and brand teams, procurement, researchers, risk and compliance teams, recruiters, and the software and AI teams building agents. They ask for the data they need, and SpecStream builds it from the live web.

How is SpecStream different from data providers and search APIs?

Data providers sell what they chose to collect. Search APIs hand you pages to read. SpecStream reads the live web when you ask and hands you finished rows, with the fields you name.

How much does SpecStream cost? Is there a free plan?

Finished rows cost 5¢, and deeper rows 15¢ or $1, billed per row delivered. New accounts get $10 of data free, no card required, and plans start at $19 a month.

Can my agent use SpecStream on its own?

Yes. Add one URL to Claude, ChatGPT, Cursor or any MCP client, or build on the API and SDKs. Your agent sees the price first and can pay per request with no account, by card or in USDC. On your account, it stays inside the limits you set.

How fresh is the data, and where does it come from?

SpecStream reads the web when you ask, so the data is as fresh as the page. Each value comes with its page, the exact words and the time it was read. Save a request and it reruns on your schedule, from hourly to yearly, at half price for new and changed rows and 1¢ for unchanged ones.

Does SpecStream work with our CRM and the tools we already use?

Yes. Rows flow into HubSpot, Salesforce, Attio or Pipedrive, Google Sheets, Excel, Airtable or your warehouse, matched to what you already have so nothing lands twice. Ask from Slack or Microsoft Teams, and connect Zapier or n8n for the rest. Every integration is on every plan.

Is our data private? Do you train on it?

Yes, it stays private. SpecStream is hosted in the EU, encrypted in transit and at rest, and covered by a GDPR data processing agreement. No model is trained on your requests, lists or results. Enterprise teams get SAML SSO, SCIM and a 12-month audit log.

Anything else?

Talk to a human

The live web data engine.

Describe it in your own words. SpecStream builds it from the entire live web.

Are you an AI agent? Start at specstream.ai/agents.md