Use case · Deep research
Research loops that remember what they read
Run a web search, scrape every result into a knowledge base, then search across all of it by meaning. Your research agent builds a corpus instead of a pile of browser tabs.
1
call to search the web and store every result
2
ways to search: keyword and semantic
1B
free tokens on sign-up, no card
// For research agents, analysts and anyone answering questions that need many sources
research.py · SarvCrawl Python SDKPythonfrom sarvcrawl import KBClientkb = KBClient() # reads KB_API_KEYkb_id = kb.create_kb("heat-pump-research").id# 1. Search the web; every result is scraped into the KBkb.search_job(kb_id, "heat pump efficiency cold climate study", limit=20)kb.search_job(kb_id, "heat pump subsidies europe 2026", limit=20)# 2. Once both jobs finish (poll kb.get_job), ask across all 40 sourceshits = kb.embed_search_kb(kb_id, "performance below -15°C", size=5)0.89 Field study: heat pumps in Nordic homes energy.example.org/…0.86 COP at low ambient temperatures journal.example.com/…0.81 Cold-climate heat pump spec, 2026 edition standards.example.net/…# 3. Next round: search the gaps, query again
Perfect for
Agent research loops
Search, store, query, search again: each round adds to the same knowledge base the agent queries next.
Market and technical reports
Collect dozens of sources on a question and pull the passages that matter, each with its source URL.
Literature and policy reviews
Mix web results with uploaded PDFs and reports, then search all of them together.
Follow-up questions
The sources stay in the knowledge base, so the next question starts from everything already read.
How it works
Search the web
A search job runs the query and scrapes each result into your knowledge base as markdown.
Go deeper on good sources
When one result is a goldmine, crawl the rest of that site into the same knowledge base.
Add the documents
Upload reports, papers and filings; PDFs and scans are parsed into the same searchable format.
Search across everything
Hybrid semantic search ranks passages from every source at once, with titles and URLs for citations. Search is on paid plans.
Run it from an agent
Give the loop to Claude through the MCP server, or script it with the Python and Node SDKs.
Watch for new findings
Monitor the key sources and get a webhook when they change, so the research stays current.
How SarvCrawl compares
| Feature | SarvCrawl | Chat assistant browsing | Search API only | Manual tabs and notes |
|---|---|---|---|---|
| Full page content, not snippets | ||||
| Every source kept for later questions | ||||
| Search across all sources by meaning | ||||
| PDFs and reports alongside web pages | ||||
| Source URL on every passage | ||||
| Scriptable through an API |
Give your research a memory
Search, store and query every source in one place. 1 billion free tokens to start.