Kimiya did not start as a language. It started as a worry. Its author — call them Φ — works where software meets trust: the study of what can go wrong in a system, who can make it go wrong, and what it takes to rely on one anyway. When language models moved out of the chat box and into the work of real decisions, that worry turned specific. People were handing these models genuine problems and getting back answers that were often right, sometimes wrong, and never quite willing to say which. The oldest question in Φ’s field returned, asked of a new kind of machine: how could you ever trust this?
The usual answers did not satisfy, so Φ did what a builder does when the theory won’t come — built things. Working systems, on real models, for real tasks: the hands-on use that teaches where a tool actually breaks rather than where you assumed it would.
The turn came from an unglamorous place. Φ was building a defensive privacy tool — meant to show a
person what an outsider could quietly infer about them from the traces they leave online, so they could
defend against it. Testing it meant comparing pieces of inferred information: is this the same as
that? And the ordinary tool every programmer reaches for — checking whether two things are
equal — simply stopped making sense. Two descriptions could mean the same thing for one
purpose and different things for another, and never “the same” in the absolute way
= demands. What was needed was not equality but a softer relation:
the same, for a purpose. That small, stubborn failure is the seed the whole language grew
from.
The larger idea arrived away from the desk, on a walk. Φ had been turning over the way people were bolting model calls onto ordinary languages — fitting the new thing into the old machine — when the inversion landed: we reach for a language model precisely to solve the problems a conventional processor cannot. So the model is not a library the old machine calls. The model is the new processor. And if that is true, then everything built around the old one — how we branch, how we loop, how we decide two things are the same, how we prove a program does what it claims — has to be rebuilt for a machine whose basic operation is meaning.
That rebuilding is Kimiya, and its through-line is a single conviction, moral as much as technical: a system should never counterfeit certainty. It should say plainly what it assumed, charge an honest price for what it cannot prove, and record its reasoning in a form a person can check without redoing the work.
The name
The name was settled on July 6th. Kīmiyā — alchemy — named, in classical usage, the means of transmutation; kīmiyā al-X meant the means of obtaining X. Al-Ghazālī’s Kīmiyā-yi Saʿādat is “the alchemy of happiness.” Kimiya is meant as a kīmiyā al-ʿaql — a means of obtaining warranted reasoning from a stochastic substrate, turning the base metal of an unreliable guess into something one can stand behind.
The homage is load-bearing, not decorative. Al-Ghazālī’s sharpest objection was to those who claimed demonstrative certainty where the conditions for demonstration could not be met — exactly the failure a discipline for machine reasoning must be built not to commit. It is why Kimiya’s guarantees are conditional, cited, and re-testable; and why, where the conditions fail, it refuses the certificate rather than fake one.
The mark
The seal is the alchemists’ oldest diagram, the squared circle — a circle within a square within a triangle within a circle, their emblem for the stone that transmutes. At its heart sits the tilde, ∼: the same, for a purpose. Read it inward: the outer circle is the certificate; the square, the decidable kernel; the triangle, the transmuting fire of the instruments; and at the center, meaning itself — measured, priced, and made dependable. The colour follows the roles: gold certifies, red is the instruments’ fire, amber the checkable kernel, and the cool teal at the centre is meaning.