Resources for journalists, analysts, and partners. One link to everything.

Werner Santos is the sole founder and architect of Werner Harmonic Labs LLC, a Woodland Hills, California–based deep-tech lab building governed-execution infrastructure for regulated AI deployment. Self-taught systems architect. 25 filed provisional patents. Currently building the most complete reference implementation of a Safe LLM Agent Runtime that exists.

Werner Oswaldo Santos founded Werner Harmonic Labs in Woodland Hills, California. A former loss-recovery agent and audio engineer turned self-taught systems architect, Werner designed the entire WHL governed-execution stack alone before using AI as an implementation crew. The architecture draws on signal processing, harmonic analysis, and formal verification — and applies them to the open problem of safely deploying LLM agents in high-stakes decision loops. Werner has filed 25 provisional patents, drafted 5 additional bundles, and built a Safe LLM Agent Runtime that ran continuously for weeks in production, producing measurable self-prediction improvement and a hash-chained audit trail of every adversarial test fired against itself. WHL operates as a sovereign lab: no venture capital, no team, no public open-source publication. Selective engagement only.

Werner Santos is the founder and architect of Werner Harmonic Labs (WHL), a Woodland Hills, California–based sovereign deep-tech lab. WHL builds the substrate for the regulated-AI deployment market: a continuously-running, self-modeling, audit-chained, adversarially-tested LLM agent runtime with bounded online learning and a measured self-prediction convergence curve — the category of system that the field is converging toward but hasn't yet assembled. Werner designed the architecture before AI coding assistance was available; AI was the construction crew that translated his designs to Python. WHL holds 25 filed provisional patents (USPTO 19/567,170) covering governance architecture, hardware-attested execution, deterministic state-machine governance, thermodynamic training loss, and hardware-bound cryptographic licensing. Five additional patent bundles (~64 new claims) have been drafted and are pending attorney review. The lab's flagship runtime ran for two production periods in March and April 2026, generating ~440 MB of structured runtime data across 13+ hash-chained ledgers. Werner's selective engagement model favors pilots with regulated enterprise, SBIR Phase III pathways with defense primes, and strategic licensing to system integrators. Background: former loss-recovery agent (regulatory exposure), audio engineer (signal processing), self-taught programmer (systems thinking). No venture capital. No team. No public open-source publication.