Consensus web search
A search API that queries Google, Bing and DuckDuckGo in parallel from Aleph Cloud's decentralized network and ranks results by cross-engine agreement.
A single POST to the search API fans out to three engines at once, executed by workers running as virtual machines on Aleph Cloud, the decentralized compute network the rest of our stack runs on, spread across North America, Europe and Asia. The responses come back deduplicated by URL and ranked by consensus: a link that all three engines return outranks the top hit of any single engine. This turns out to be a useful spam filter, because SEO manipulation that fools one ranking algorithm rarely fools all three at the same time.
The API serves web, news, image and academic search, plus a fetch endpoint that returns clean, readable page text for any URL. It runs on the same credits as our inference API, and agents can alternatively pay per query over x402 micropayments without creating an account.
Privacy properties
Queries and page fetches reach the web from the decentralized network rather than from your machine, so nothing ties a search to you, and there is no central datacenter accumulating everyone’s queries in one place. We keep no search history attached to a person and build no interest profiles. This matters most for agents, which tend to search with far more context about their owner’s intentions than a human typing keywords would reveal.
Pricing
A single-engine query costs $0.002 with DuckDuckGo or $0.003 with Google or Bing, which works out to $2 to $3 per 1,000 queries. Full three-engine consensus mode is $8 per 1,000. For comparison, in June 2026 Tavily charged $8 per 1,000 searches, Exa $5, and Brave $4 plus token fees, each as a separate vendor with its own account and billing.
Where it shows up
The CLI exposes it as libertai search and the MCP server exposes it as a web_search tool, so coding agents and chat clients can use the same consensus results.