Problem

Cajal addresses the trust gap created when AI generates software faster and more complexly than humans can audit. Its site argues that AI systems can find flaws human reviewers miss, while AI-written code may exceed the limits of human comprehension; testing only checks selected inputs, whereas a mathematical proof can establish behavior across every possible input. The economic pain is therefore a combination of scarce, expensive expert review and the potentially severe cost of undetected defects in mission-critical software. The clearest “killer” use case is making AI-generated or otherwise critical binaries provably correct, although the available evidence does not quantify the savings or cost of failures.

Product / Service

Cajal’s flagship product is Tau, a reasoner and prover that works directly on a compiled binary. It captures the customer’s intent as formal specifications, proves those specifications mathematically, and returns either an audit-ready certificate of correctness or a report identifying bugs that invalidate them. Tau is built on Talos, Cajal’s open-source interpreter for lifting binaries into a form that can be reasoned about formally. The benefit is machine-checkable assurance without requiring a human to understand every detail of complex code.

The company also describes a broader AI-mathematician offering: Tau is a multi-agent system that discovers and verifies proofs in Lean and autonomously formalizes applied mathematics, initially targeting quantum computing and finance. Cajal appears to combine product-led enterprise work, reached through a one-to-one demo, with collaborations for frontier AI labs and research institutes involving datasets, evaluations, and reinforcement-learning environments. Its terms target commercial entities and research institutions, and provide for paid features or order forms, but do not disclose pricing.

Market

Cajal competes in formal verification and formal methods, AI-assisted theorem proving, and developer infrastructure for trustworthy software and mathematical reasoning. Lean is the open-source proof-assistant ecosystem on which Cajal says its applied-mathematics work is built. Adjacent competitors and alternatives include Harmonic’s Aristotle, Logical Intelligence, DeepSeek-Prover-V2, and formal-verification providers such as Galois, Theorem, Axiomise, Certora, and CertiK; a broader directory also names Hugging Face, Microsoft, and GitHub as possible alternatives. These should be viewed as a mixed set of direct, adjacent, and ecosystem competitors rather than a confirmed list of head-to-head rivals.

Public traction is early-stage rather than demonstrated at scale. Cajal was founded in 2025, entered Y Combinator’s Winter 2026 batch, and is listed as active with a two-person team. The company is soliciting demos and research collaborations, while its legal terms establish a mechanism for paid services; however, the available record reports no revenue, customer count, or named public customers. For diligence purposes, Cajal is best characterized as pre-scale and effectively pre-revenue in the public record, with commercial intent but unverified commercial traction.