Managed API
Send queries to us. Nothing is kept unless you turn on learning.
Innovation · search packsIn development
A search engine you can actually own.
Search is getting locked away. Bing’s search API is gone, and Google’s stops searching the whole web on January 1, 2027. So we’re building the alternative: the web’s knowledge as packs you run in your cloud, your building, your app or ours, with a receipt on every result.
A result your agent can cite
Searchable by keyword and by entity, from a server your agents already know how to call.
- Source
- Where the text came from
- Snapshot
- The date of the copy it read
- Fetched
- When it was fetched
- Why it ranked
- The reason it came up first
Five sizes
From your pocket to the whole web.
The smallest pack already ships inside Corollary. The rest grow from there, on the same engine.
Pocketabout 0.25 GB
30,000 subjects indexed, 530,000 more archived
A phoneInside Corollary today
Knowledge0.4 to 3 GB
The top ~377,000 subjects: 80% of what people read on Wikipedia
A phone, laptop or small containerIn development
Encyclopedia10 to 60 GB
All 7.2 million English article subjects, and their text
A laptop or containerIn development
Web Sabout 1 TB
All of that, plus the top few hundred million web pages
One machine with fast local diskIn development
Web L10 TB and up
Billions of pages, rebuilt monthly, updated weekly
A small clusterIn development
Four ways to run it
It runs wherever you do.
One build, the same index and the same answers. Only where it runs changes.
Marketplace, self-hosted
It runs inside your own account, and queries never leave it.
Licensed on premises
Your servers, any cloud, even networks with no internet at all.
On-device license
It ships in your app, on each person’s phone or laptop, and questions stay on the device.
The learning option
It learns. Only for you.
Turn it on and the index fits itself to you, through a short, fixed list of adjustments. Every one lands on a receipt you can read, undo or export. Your layer never flows to anyone else.
Misses grow itA question it can’t answer fetches the missing subject from a public source, and it joins your layer. On a device, the fetch asks for a whole bucket of subjects, so no request names what was asked.
Your documents joinYour own files are indexed right beside the web, with the same receipts.
Ranking fits your useSources your agents cite rise. Sources you block drop out. It learns from receipts, never from reading your words.
Who it’s for
For questions that can’t leave the building.
Banks, insurers, hospitals.
Agents that need current outside information, where queries may not leave the network and every answer needs an audit trail.
Self-hosted, with receiptsLife after the big search APIs.
Bing’s API is gone and Google’s whole-web search ends in 2027. Point your agents at a search you control.
Managed API or your AWS accountWorld knowledge, on the device.
Small on-device models know little about the world. Ship a pack in your app and answers stay on the phone.
On-device licenseWe’re picking our first pilots.
Building agents or apps that need the web without giving away the question? Tell us what you’d run, and where.