We are building a planet that machines can read.
SOLYZER builds Open Earth: a live digital twin of the world, assembled from open data and served as one contract that humans, AI agents, and eventually robots read to decide. Not a map to look at, a world model to act on.
Perception should not be rebuilt by every agent.
Today every system that needs to understand a place starts from nothing. It burns tokens and reasoning cycles rediscovering what is already knowable: what stands there, how large it is, how busy the street is, what is flooding nearby.
That is waste, and worse, it is unaccountable. Each rebuild invents its own confidence and forgets where the number came from. Perception is not a per-agent problem. It should be a utility an agent calls instead of reinventing.
From open data to a defensible belief
Read the ground
Places, roads, buildings, traffic, hazards, vessels, flights, satellites and street-level imagery, pulled from open sources that permit it.
Cross-check it
A reconstruction is a hypothesis until independent references agree. Where they disagree, both readings survive and the contradiction is recorded.
Price the uncertainty
Every figure carries its confidence and its caveats. When the evidence is too thin to answer, the answer is a stated unknown, not a number.
Serve it to anything
One machine-readable contract over REST and MCP, so a person, an agent, or a future embodied system reads the same ground the same way.
Rules we hold even when they cost us
A world model is only worth what you can trust of it. These hold on every response, and they are checkable rather than promised.
Unknown is never zero
If a source could not answer, we say so. Treating silence as absence is the most common way spatial data lies.
Caveats travel with the number
Every brief ships the assumptions and sampling limits that produced it, inside the response, not in a footnote nobody reads.
We refuse to invent
A null score means the system declined to guess. Imputing a plausible number is how a dataset becomes confidently wrong.
Licenses stay separable
Share-alike and permissive sources are labelled per record, because the obligation you inherit depends on which one you used.
The full contract is published for machines in the documentation, and an agent reads it before it reads a number.
License-clean, or it does not ship
The twin is built from sources that permit it. We refuse to scrape data we cannot serve onward, and we keep each record's license attached so anyone downstream inherits the truth about what they are allowed to do with it.
- 27
- data sources catalogued
- 16
- need no API key at all
- 15
- we are allowed to serve onward
We score what we build, in public
A perception layer is only worth what gets decided on it. So what we build on top publishes its results the same way the data does: recorded before the outcome exists, then graded against reality.
The Alpha Engine, our on-chain use case, writes down every call with its entry, targets and stop before anyone knows how it resolves, then scores it on real prices. 2,381 calls have resolved, and every one is published, losses included.
Powered by Hudhudbird Perceptrons
The twin's perception is not a black box. It runs on Hudhudbird Perceptrons, an open framework that forms a spatial hypothesis, cross-checks it against independent references, and prices the uncertainty into every figure. Hudhudbird builds perception. Solyzer builds the world.