We got tired of renting intelligence.
Every other AI company sells access.
Access can be revoked, repriced, throttled, surveilled,
or deprecated at any time.
You are renting intelligence from someone who can
change the terms whenever they want.
SimpleDirect sells the asset itself.
You download the model. It runs on your machine.
Your data never leaves your network.
No per-token billing, no terms of service
that update quarterly.
Once you have it, it is yours —
permanently, unconditionally, and irrevocably.
What SimpleDirect is
We build specialized open-source Canadian Business + Legal + Regulatory AI you can own. There are four ways to use our models:
Open weights, free.
Download flash-1-mini from Hugging Face under Apache 2.0 and run it anywhere. The weights are yours.
Free showcase API.
A browser way to try the models on Canadian sovereign compute — free account, generous quotas. Launching September 30, 2026 at chat.getsimpledirect.com.
Licensed cloud-deploy package.
Deploy SimpleDirect in your own AWS / Azure / GCP tenancy — models, RAG infrastructure, corpus, and a forked Open WebUI. Your data stays in your tenancy.
Future on-prem appliance.
A hardware-bundled, plug-and-play unit for regulated buyers who cannot use the cloud. Exploration for 2027+.
None of these is a subscription. None is SaaS that requires you to send us your data.
What SimpleDirect is not
Not a SaaS business.
We don't run a hosted product where buyers send us their data. The free showcase exists so people can try the models — it isn't the thing we build a sales organization around.
Not a subscription company.
The open weights are free forever. The showcase has free quotas. The cloud-deploy package is a license, not a subscription.
Not competing with frontier labs.
We are not building GPT-5. We are building the best specialized model for Canadian Business + Legal + Regulatory work that buyers can actually run, own, and trust on their own hardware or in their own cloud.
Not a data company.
For self-deployed customers, your data never reaches us. For showcase users, the transparency disclosure at /data explains exactly what is and isn't stored. We don't train on your conversations.
Not a consulting firm.
We ship software. George writes about AI ownership. Those are separate.
How we decide
Every product decision gets filtered through one question:
"Does this make it easier for someone to own their AI?"
If yes, do it. If no, kill it.
Model update notifications? Yes
Helps you keep your owned model current.
Free showcase API on Canadian sovereign compute? Yes
Most buyers can't deploy from Hugging Face. The showcase lets anyone try the model in a browser — transparent data handling, free quotas, no required upgrade path.
Paid API tier as the primary business? No
We're not running a sales motion for API access. If the free showcase gets popular and people want pay-as-you-go for higher quotas, we'll consider it then — not before.
Licensed cloud-deploy package for production? Yes
Buyers deploy SimpleDirect in their own AWS / Azure / GCP tenancy. Their data stays in their tenancy. We don't host it.
Releasing training datasets? Yes
Yes — but after the specialization edge is preserved. We hold a series' corpus during active training of related models; the flash corpus releases with flash-1-pro in March 2027.
Subscription model? No
The cloud-deploy license is annual usage rights, not a vendor-hosted service that cuts off if the license lapses. The deployment you've built keeps running.
Adding telemetry? No
Violates the ownership promise. Permanently.
Per-seat licensing? No
Ownership is not per-seat. It is per-organization.
Sales team? No
We respond to inbound. Deployments and engagements arrive through content authority and existing relationships. No outbound sales motion.
Team
George Pu
Founder
Founder of SimpleDirect®. Building toward AI you can own — open-weight Canadian Business + Legal + Regulatory models that run on your hardware, a licensed cloud-deploy package for production use, and a future on-prem appliance for regulated buyers who can't use the cloud.
Writes about Canadian AI infrastructure decisions at Founder Reality and posts daily on X.
SimpleDirect® trains models on Canadian sovereign AI infrastructure in Rimouski, Quebec. Every analysis published is grounded in infrastructure George actually runs.
Vikas Grover
COO & CFO
COO and CFO at SimpleDirect. Runs operations, finance, partnerships, and the analytics that keep the company honest about what's working and what isn't.
Five years operating Canadian startups. Aeronautical Engineering degree from Toronto Metropolitan University. Formerly a commercial pilot at Air Canada.
Ayush Naik
Technical Lead
Principal Systems Engineer at SimpleDirect, focused on the infrastructure and systems beneath the surface — the foundation that determines whether anything above it stands.
A constant tinkerer who'd rather test new tech than read about it. Found outdoors when he can be, with a book somewhere within reach. Picks up a pencil now and then.
Engineering team based in Canada and around the world.
How we operate
Operator-credible, not consultant-credible.
Every analysis we publish is grounded in infrastructure we actually run. We build what we recommend. We deploy what we build.
Lean by design, not by accident.
We choose to stay small so the principal stays in the work. The depth comes from focus, not from headcount.
Content and capability compound.
Public analysis on AI infrastructure decisions and shipped capability — open-weight models, deployment toolkits, reference architectures — are the dual mechanism. Both intentionally visible, intentionally documented, intentionally reproducible.
Canadian by structure, sovereign by operation.
The work originates in Canada, runs on Canadian sovereign infrastructure, and is governed by Canadian law. Design choices that matter — not marketing posture.
Public process, durable record.
We publish what we learn. Access to Information requests, blog posts, model releases. The body of work is the credibility, not the org chart.
No tools without need, no platforms without proof.
We do not build software people haven't asked for. We do not position as a platform before there is evidence. The discipline is the moat.
What to expect
Engagement is selective.
Two to three Tier 3 engagements per year maximum, by inbound only. Tier 3 engagements are scoped technical deliverables for specific high-value buyers — federal departments, top law firms, regulated enterprise. We do not run an outbound sales pipeline. We do not pitch new business.
We respond to inbound within five business days.
Including saying no when the fit is wrong. We say no with reasons we can explain.
Engagement scope is defined upfront.
We do not take open-ended retainers. We do not take "let's figure it out as we go" engagements. We define what we will deliver before work begins.
We publish what we learn.
With client approval, anonymized where required, we write up the substance of completed engagements. The work compounds into the body of work.
We tell you when you don't need us.
Many organizations asking about sovereign AI infrastructure already have the right answer (commercial API with a Data Processing Agreement). We will say so when it's true.
Where we are on the ladder
We are climbing the same ladder we write about.
Right now we train models on Canadian sovereign AI infrastructure in Rimouski, Quebec.
Later this year, colocation. Our own hardware. Fifty steps from the office.
We document every step — real costs, real tradeoffs, real experience. Because nobody else is doing that for people who don't already speak the language.