Private AI for law firms. Client files stay in the building.
We design, install, and run firm-wide AI infrastructure inside your offices or data center. Every practice group gets near-frontier models for review, research, and drafting, and not one document goes to a cloud provider that could be compelled to hand it over.
This page is general information, not legal advice.
Every prompt is a disclosure decision.
- When a lawyer pastes client information into a cloud AI tool, that information now lives on someone else's servers, under someone else's terms. Privacy settings help. They don't change the fact that a provider holding your data can be breached, subpoenaed, or legally required to produce what passed through its systems. The ABA's Formal Opinion 512, issued July 29, 2024, makes evaluating that exposure your job, not the vendor's. An Opichi solution puts you in charge of controlling exposure.
Six duties. One question:
Where does the data go?
Opinion 512 doesn't ban generative AI. It applies the Model Rules you already follow to a tool that, by default, sends your work somewhere else.
Understand what the tool can and can't do, keep up as it changes, and independently verify its output before you rely on it.
Assess the risk of disclosure before any client information goes in. For self-learning tools, the client's informed consent comes first, and a boilerplate clause in the engagement letter isn't enough.
Tell clients when AI touches their information, your fee, or a significant decision in the matter, and whenever they ask.
Check every citation, analysis, and argument before it reaches a court. A hallucinated case is still your misstatement.
Set firm policy, train lawyers and staff, and vet any outside AI provider: its security, retention, breach notification, and claims on your data.
Bill actual time, treat general tooling as overhead, and pass through only reasonable costs the client knows about.
Take the third party out of the equation.
- Most of Opinion 512's hardest questions start with sending client data to someone else. A private model removes that step. We've been building software for businesses since 2014. Today much of that work is private AI for organizations with too much at stake for the public cloud: the servers, the models, and the safeguards, built inside your own walls.
CLOUD GENERATIVE AI
・ Client data is processed on the provider's servers・ The provider can be compelled to produce your prompts・ You must vet terms of use, privacy policies, retention, and breach notices・ Does not work without the internet・ Per-seat or per-use cost, rising with adoption・ Frontier models for fast speed and capability・ Subject to prompt misinterpretation, hallucinations, and the AI company's main system prompts
PRIVATE AI WITH OPICHI
- ・ Client data is processed on your local network, on servers your firm owns
- ・ You have complete ownership over your prompts
- ・ No terms to vet; you're covering your data completely
- ・ Private AI can run fully isolated
- ・ Cost is in the infrastructure you own, plus design, build, and support
- ・ Near-frontier, tuned to your balance of speed and quality
- ・ Main system prompts tuned to your specific requirements, significant reduction in hallucinations, and easily verifiable data
Infrastructure for the whole firm, not a desk.
- Hundreds of attorneys, every practice group, whole data rooms at once. We engineer the system around how your firm actually works. We size compute, models, and storage to your headcount, document volume, and practice mix, then tune the balance of answer quality and speed for each kind of work.
Capacity
Server-grade GPU infrastructure in your offices or data center, sized so many attorneys can work at once without queuing behind each other.
Volume
Full-document analysis across large sets: due diligence, discovery review, contract portfolios, and research memos, with answers tied to the sections they came from.
Model mix
Larger models where depth matters, faster ones where turnaround does, each matched to the task rather than one model for everything.
Matter isolation
Permissions by matter, practice group, and role, so information from one representation stays out of another.
Your rules
Firm playbooks, clause standards, and review checklists run with every prompt, so the system checks what your partners care about.
Growth
Add capacity as adoption grows, across offices, without renegotiating seats or sending more data out.
From first conversation to firm-wide rollout.
Map the work and the risk
We meet with practice leaders, IT, and your general counsel to map the work, the sensitivity of the files, and what your ethics obligations require.
Architect the system
We size compute, models, and storage to your attorney count and document volume, and design access around matters, groups, and ethical walls.
Install and lock it down
We install the infrastructure in your offices or data center, configure full document analysis and your review rules, and pass your security review.
Put it to work
We pilot with one group, train attorneys and staff, document the system for your AI policy, and expand as adoption grows.
What firms ask us first.
Bring the work you'd never paste into a chatbot. We'll run it on infrastructure you own.
Fifteen minutes with a senior engineer. You'll leave with a clear view of what private AI at your firm would take, and what's worth building first.
Talk to our team