AI that earns
its keep.
Onyx AI Labs is a lab. We build AI systems, train our own models, and publish what we learn. Some of it becomes a product; all of it gets tested before we say so.
Built, shipped, running
Production systems we built and operate ourselves. Every status badge is literal.
What we do
Three things, done properly. No transformation theater.
Deploy
Working AI in your business — workflows, knowledge assistants, agents — with permissions, data boundaries, and human approval built in from the first commit. Running systems, not slide decks.
Govern
Policy, identity, audit trails, and decision receipts for AI you already run — on OpenAI, Anthropic, Gemini, or your own hardware. Enterprise controls without the enterprise overhead.
Advise
Eighteen years of automation across regulated industries, and seven of them spent on AI that had to be defensible. We find the opportunities that pay, skip the traps, and write a plan your team can actually execute.
Enterprise rigor.
Delivered at pace.
We deploy, not just advise
You get running systems, not slide decks.
Multi-model, not locked in
We use the right AI for each job — Claude, GPT, Gemini, open-source, or custom.
Controls from day one
Approvals, audit trails, data separation — built in, not bolted on.
How we work
Discover
We learn your operations, identify the highest-impact AI opportunities, and define a clear scope.
Design
Architecture, model selection, integration points, and controls — mapped before we write a line of code.
Deploy
We build and ship working systems into your environment. Not a demo — production.
Improve
Monitor, measure, and iterate. AI systems get better with use — we make sure yours does.
AI that answers to you.
Not to a rate limit.
Sovereign AI means the model is yours to run. It isn't a checkbox — it's an engineering problem with a real answer, and it's the problem we spend most of our time on.
Your models, your weights, your data — running on infrastructure you control. Nothing leaves the building that you didn't choose to send, and no vendor can revoke access, change a price, or retire a model out from under you.
Full capability without the dependency: inference at the speed your workflows need, private fine-tuning on your own documents, and an audit trail you hold. The same results as a hosted API, minus the third party in the middle.
Real engineering. Model selection and quantization that fits your hardware budget, inference serving tuned for your concurrency, evals that prove the local model is good enough, and an operations plan for the day a GPU fails.
Sovereign AI doesn't need a datacenter. It needs a box with enough memory to hold the model — that's the whole constraint. A DGX Spark or a Mac Studio with unified memory runs a serious model at useful speed on a desk. A single workstation GPU covers smaller models and fine-tuning. A rack of them, or a small cluster over a fast interconnect, covers the big open-weight models at production throughput. We size it to the model you actually need, not to a vendor's spec sheet.
We'll tell you what your workload needs before you buy anything — and we'll tell you when the honest answer is “you don't need this yet.”
A lab, not a pitch deck.
We build it, run it, and show the numbers.
Onyx AI Labs was founded by C. Michael Gibbs to build AI systems that hold up under real workloads — with models we train ourselves, measured against benchmarks we publish. We'd rather ship something and show you the throughput than describe it in a deck.
Based in Toronto, with offices in New York and UAE (opening 2027). Delivered across North America, Europe, and Asia-Pacific.
Let's build something.
Whether you're exploring AI for the first time or ready to deploy, we'd like to hear from you. Start with a working session — no pitch decks, no pressure.
Start a working sessionToronto / New York / UAE (2027)