yforest AI Labs
Trustworthy Local AI
yforest AI Labs develops the measurement science, systems, and cryptographic tooling to make on-device artificial intelligence provably private and measurably energy-efficient.
The problem
America is scaling AI through centralized datacenters — concentrating energy demand into gigawatt-scale facilities and routing sensitive data through third-party cloud APIs. For law firms, schools, clinics, and government offices, that architecture is a compliance non-starter, and for the grid, it is a growing burden.
Meanwhile, hundreds of millions of efficient AI-capable devices already sit on American desks. Our research asks a simple question with hard technical depth: can the AI workloads that matter run entirely on hardware you already own — with cryptographic proof that your data never left, at an energy cost that beats the datacenter?
flagship projectAEOLI — Attested Energy-Optimal Local Inference
The EQAT Benchmark
An open methodology for comparing the true end-to-end energy cost of AI inference — datacenter versus device — normalized by output quality. Because today, nobody can even measure this honestly.
Energy-Optimal Runtime
Scheduling AI inference across CPU, GPU, and Neural Engine on unified-memory consumer silicon, with dynamic quantization and memory management tuned to the hardware Americans already own.
Verifiable Data Locality
Hardware-backed attestation that an AI session ran entirely on-device with zero data egress — turning "trust me, it's private" into an artifact a compliance officer can independently verify.
Results, specifications, and the benchmark harness will be published openly.
Built on shipped systems
Among the earliest complete macOS applications running modern LLMs fully locally on Apple's Metal architecture.
An AI ETL/IDE that relocates enterprise data-transformation workloads from cloud infrastructure to local Apple silicon.
A published Chrome extension that strips PII from prompts before they reach cloud AI services.
Privacy-first consumer applications in education and personal health. anthosair.com
Research library
Evidence reviews, measurements, and open studies from the lab. We publish what we find, including when it's inconvenient for our thesis.
Nobody Can Honestly Compare Cloud and Local AI Energy
The four measurement problems behind the EQAT benchmark.
evidence reviewThe Physical Footprint of Centralized AI
Energy, water, and land — in numbers, from federal laboratory and congressional sources.
evidence reviewThe Orbit Question
What space data centers tell us about terrestrial limits.
evidence reviewThe Catch-Up
How fast local models are closing the gap — and the caveats we insist on.
evidence reviewThe Architect and the Workers
Most daily AI tasks don't need a frontier model. The routing literature, quantified.
evidence reviewCan the Machines on Our Desks Actually Do This?
Hardware readiness, the five gaps, and the grid dividend stated conservatively.
evidence reviewThe Compliance Wall
Why regulated professions legally cannot simply use cloud AI — HIPAA, FERPA, privilege, and 35+ state bar opinions.
evidence reviewPrior Art and the Competitive Landscape
What llama.cpp, MLX, Ollama, and Apple's Private Cloud Compute already do — and what none of them provide.
evidence reviewThe Texas Grid and the Demand-Side Question
ERCOT's 410 GW queue, SB 6, and the water projections — the debate as it's happening in our own state.
evidence reviewWhen the Data Is Too Sensitive to Send
The first agentic production intrusion, the forensic wall its defenders hit, and what it proves about data that cannot leave.
planned studyThe Local AI Adoption Study
Prospectus for the first longitudinal study of local AI adoption. Seeking participants.
Who is doing this work
founder · yforest ai labs
Christopher M. Davidson
Christopher M. Davidson is the founder of yforest AI Labs and leads its research program. In his day role he serves as Vice President of AI Engineering for the Dallas Mavericks, where he has spent eight years and helped found the organization's strategy and analytics function — building the data architecture that the business runs on.
His applied work is the evidence base for this research. He has designed and shipped eight production systems, including yforest Analyst — among the earliest complete macOS applications running modern LLMs entirely on-device via Apple's Metal architecture — QueryFlow, an AI ETL/IDE that moves enterprise data transformation off cloud infrastructure and onto Apple silicon, and Cloak, a published extension that strips personal information from prompts before they reach cloud AI services. The full portfolio is documented in the journey.
That applied work is why the lab exists. Building production systems on-device surfaced the same unanswered questions again and again — how to measure the real energy cost, how to orchestrate inference across a machine's hardware, how to prove to a regulator that data never left it. Those became AEOLI's three research thrusts. The same team ships commercial work under yforest AI Labs: automation and lead systems for growing businesses, enterprise data tooling like the recently launched QueryFlow, and Build With Me, where clients build their own working AI system alongside us and keep the skills.
Work with us
Pilot partners
We work with organizations whose documents cannot leave their control — legal, education, healthcare, government.
pilot inquiry →Research interns
Texas-based students in systems, ML, or security.
research internship →Collaborators & advisors
Academics and practitioners in measurement science, trusted computing, and efficient inference.
collaboration →