They search the web. We search social.
Universal Search runs one natural-language query across up to 17 social sources at once. An AI plans and prunes before anything runs, results fuse so a post seen on many platforms rises, and a second AI agent reranks on real comments and transcripts, then clusters into themes. Streamed live.
Crawling data from the world's biggest platforms
TikTok
Instagram
YouTube
X
LinkedIn
One query in. One ranked, clustered answer out.
Watch a single question move through the pipeline: planned, pruned, fanned out across every platform at once, fused, reranked on what people actually said, and clustered into themes.
what are people saying about ___?
AI plan
Intent and sub-queries. A deterministic plan ships in under 50ms.
AI prune
Drops off-topic sources before they bill.
Parallel fan-out
Every relevant source fires at once.
14 platforms · up to 17 sources
RRF fuse
k=60. Ranks merge across platforms.
The same post seen on four platforms rises.
AI rerank
Weighted on real comments and transcripts.
Cluster
Grouped into themes.
Ranked, clustered answer
How does one query search 17 places at once?
Three moves turn a natural-language question into a ranked, clustered answer.
An AI reads intent, then trims the search
It expands your question into sub-queries and scores each source's fit. Sources that won't help are dropped before they cost a credit.
Up to 17 sources, fired in parallel
Every relevant platform is queried at once. Weighted Reciprocal Rank Fusion (k=60) merges them, so a post seen on several platforms rises to the top.
Ranked on what people actually said
A second AI agent reranks on real comments and on-camera transcripts, penalizes off-topic results, and groups the rest into themes. Streamed live over SSE.
Every number here is a real pipeline setting, not a marketing figure. See the docs for the full algorithm.
No single trick. The whole pipeline.
Every algorithm here is public. The advantage is the integration: planning, pruning, cross-source fusion, and grounded reranking, all streaming with correct billing on partial failures.
PLAN + PRUNE
A deterministic plan races two LLMs, then affinity pruning aborts weak sources before they bill. Speed and cost, handled at the top.
CROSS-SOURCE FUSION
Weighted RRF over heterogeneous social data merges the same post across platforms, so provenance becomes a ranking signal no single index can produce.
GROUNDED RERANK
The reranker reads real comments and on-camera transcripts, and penalizes results that never mention what you asked about.
HONEST BILLING
Results stream as they resolve, and a search that comes back empty is refunded. Partial failures never corrupt the response.
“The moat is the integration. Months of work to assemble, in a category web search does not serve.”
They search the web. We search social.
Exa, Tavily, and Firecrawl are excellent at the open web. Universal Search is a different data product: the live conversation on social.
| How they compare | SocialCrawl | Open-web search |
|---|---|---|
| What it searches | Social platforms, forums, and prediction markets | Public web pages and articles |
| What a result is | A real post, comment, or video, with engagement | A web page or document |
| How results rank | Engagement, freshness, and cross-platform provenance | Links, text relevance, and page authority |
| Multi-platform provenance | Yes, the same story across platforms rises | No, one page per result |
| Output | Ranked, clustered, tagged by source, JSON or SSE | A list of links or scraped text |
What it searches
What a result is
How results rank
Multi-platform provenance
Output
Ask anything. Watch it stream.
Run a real universal social search. Results arrive as each source resolves, over a live SSE stream.
Free to try, one search a day. The developer API is a flat 20 credits per call.
Every one of those sources returns a different shape.
Reddit, TikTok, YouTube, and the rest each speak their own format. Before anything is ranked, every result is normalized into one canonical schema, validated at the wire, with a deterministic block for engagement rate, language, category, and reach.
See how the unified schema worksOne call. Every platform.
Numbers reflect the Universal Search fan-out. The 20-credit price applies to the developer API; the web search is free to try.
Frequently asked questions
Can't find what you're looking for? Talk to our team or ask the AI agent below
Search social like it's one place.
Run a universal social search in the browser, or call one endpoint from your code. Same pipeline, same ranked and clustered answer.
curl https://www.socialcrawl.dev/v1/search/everywhere \
-G --data-urlencode "query=what are people saying about ___?" \
-H "x-api-key: $SC_KEY" \
-H "Accept: text/event-stream"