Private AI deployment for regulated firms
Omni Studio builds private AI deployments for healthcare, legal, finance, insurance, real estate, and government contracting firms. We install Omni Studio and its AI employees in your own AWS or Azure account, with models under no-training or BAA terms you sign. AI employees read and draft, and anything that sends, signs, or changes records waits for your approval.
Who it's for
Your client contracts are often stricter than the law. No law requires AI to run on your own hardware, and HIPAA allows cloud AI under a business associate agreement (HHS OCR, October 2016). The FDIC's outside counsel deskbook bars sensitive FDIC information from any generative AI tool in an "open environment" (§1.17, April 2026 edition).
Staff use whatever chat tool is open: among independent insurance agencies that use AI, 45% use public AI tools, and 55% have no written AI policy (IIABA, February 19, 2026).
A check you can run today: search your client agreements, outside counsel guidelines, and MLS rulebook for "artificial intelligence" and "generative." A clause that names where AI may run is the reason for a private build.
We build for firms of about 20 to 150 people and for defense suppliers. In your build, AI employee Mia drafts intake notes from new inquiries and AI employee Sam flags filing deadlines to a person, on the model in your account.
Three ways to run it
Private cloud
Most firms start here. Omni Studio runs in your AWS or Azure account, on Amazon Bedrock or Azure OpenAI (Amazon's and Microsoft's hosted model services) under terms you sign. Records, files, logs, and backups stay in your account, under your own encryption keys.
Hybrid, built per client
Private cloud, plus a model on your own hardware for records you flag as sensitive, such as Part 2 substance use records or tax return data. A routing rule can miss a record, so we test it on your files before go-live and you approve the rules.
Fully offline, built per client
Omni Studio, the model, email, and backups run on your network with no internet connection, for defense suppliers with CUI or ITAR data. Model quality is lower: the best open model scores 1746 against 1866 for the top closed model on GDPval-AA office tasks (Artificial Analysis, October 21, 2026).
What stays inside your environment
| What | Private cloud | Hybrid | Fully offline |
|---|---|---|---|
| Omni Studio, records, files, and logs | Your cloud account | Your cloud account | Your network |
| Backups | Your account, your keys | Your account, your keys | Your network |
| Model calls on everyday records | Bedrock or Azure OpenAI | Bedrock or Azure OpenAI | A model on your hardware |
| Records you flag as sensitive | Bedrock or Azure OpenAI | A model on your hardware | A model on your hardware |
| Phone calls and texts | Twilio (the phone and text service) and our voice provider | Twilio and our voice provider | A phone carrier, unless we build an offline phone system |
| Tools you connect, such as Gmail or QuickBooks | Where you choose | Where you choose | Switched off |
How an engagement runs
- We read your contracts with you and pick one workflow, for example intake notes drafted for attorney review.
- We agree one metric and a kill criterion in writing, and we propose the kill criterion first. Example: every accepted draft cites its source record, and no record reaches a provider outside your scope.
- We build in your environment and run the workflow with your staff for 60 to 90 days.
- At the check you keep it, change it, or stop. If it misses the kill criterion, we recommend stopping.
We work this way because 95% of generative AI pilots showed no measurable value in MIT NANDA's study (Fortune, August 18, 2025). Pricing is scoped with you, from the Custom Build plan, or Enterprise by quote. See pricing.
What we are not
- We hold no DoD or agency authorization, and we claim none for CUI.
- We have no SOC 2, ISO 27001, or HITRUST report yet. SOC 2 and ISO 27001 are in progress, and we make no HIPAA compliance claim (Security, October 21, 2026).
- We accept no clinical or legal liability for drafts your staff approve.
- We promise no savings. At office volume a private build usually costs more than cloud AI.
- Nothing we build finalizes a note, filing, or decision without a person signing off.
Questions
Do I need a private build to use AI under HIPAA?
No. HIPAA allows cloud services of any configuration when the vendor signs a BAA (HHS OCR, October 2016). A private build fits when a contract, license, or record type asks for more, such as Part 2 records.
Will you sign a BAA?
Not today. If you need us to sign a business associate agreement before work starts, we are not a fit yet. In a private cloud build, the model runs under the BAA you sign with AWS or Microsoft.
How long does a build take?
The first workflow runs for 60 to 90 days against the metric we agree. Hybrid and fully offline builds take longer, because the local model and routing rules come first.
Who maintains it?
The written scope says who applies security patches, who refreshes the model, and how often we review the workflow with you.
This is general information, not legal advice.