Knowledge without the cutoff
Fetch knowledge, not markdown. Diffbot grounds your AI on verified facts from the public web.
Prompt: Research "Prometheus" and draft a sales brief.
- The transcript appears here — press RUN AGENT.
Run the agent to watch the suite work — each step is a real Diffbot API call you can inspect.
Introducing web search that runs on your hardware
The dot com days are over, but web search is still monetized on ads and scraped for API access. At Diffbot, we have a better solution — Web search you own. No ads. No telemetry. It turns out, you don't need all 150TB of the web to power a competent purpose-built local search engine. We've packaged it down to just 4TB of essential news, docs, homepages, recipes, and whatever else you need.
Take an entire web index with you on a boat trip and stay online while offline.
Diffbot Web Search API currently returns responses in <300ms P90 latency while matching NDCG of equivalent "fast" modes on other web search providers. Should you choose to self-host, expect a 20-100ms latency improvement depending on hardware and shard optimizations.
Build with knowledge
Browse the rest of the tools we built to structure the world’s knowledge.
Extract a page
Point Extract at any URL. Computer-vision models classify the page, render it, and hand back clean, structured fields in about 300ms — no rules, no scrapers.
Crawl every page
Give Crawl a seed URL and it walks an entire site, turning thousands of pages into one structured dataset you can query or export.
Structure natural language
Send raw text to the Natural Language API for entities, relationships, facts, and sentiment — each one resolved against the Knowledge Graph.
Query the Knowledge Graph
Search the largest structured database of the public web — billions of people, organizations, products, and articles — like one big table.
Search the web
Run live, structured web searches you own and control, and feed fresh results straight into your agents and pipelines.
