The AI Chat App Your Firm Can Actually Point at Client Data
Every CPA firm has had some version of this conversation by now: a staff member wants to use ChatGPT or Claude to speed up a review, and someone has to say no, because the data can't leave the firm's four walls. That's not caution for its own sake. It's the correct call, because when you paste a client's trial balance or payroll file into a consumer AI tool, you have no real control over where that data goes or how it's used.
Nera's ChatApp exists to remove that tradeoff. It's a chat interface, built the way your team already expects to work with AI, that runs on top of Claude, Gemini, or ChatGPT. The difference is what happens to the data underneath. Client files stay encrypted the entire time the model is working with them. Not encrypted in transit and decrypted on arrival, the way most "secure AI" tools work; encrypted all the way through the computation. The model never sees plaintext, and neither does Nera. Only your firm, holding its own key, can decrypt the results.
That's a meaningfully different promise than what the big AI vendors already offer. Anthropic, OpenAI, and Google will all tell you, correctly, that they don't train on your enterprise data. But training was never the whole risk. The moment client financials sit somewhere as plaintext, even briefly, even behind a vendor's promise, they're exposed to everything that can go wrong with any system holding sensitive data: a breach, an employee who looks at something they shouldn't, a misconfigured permission, a subpoena or discovery request that reaches the vendor instead of you. None of that requires bad faith on the vendor's part. It just requires the data to exist somewhere in readable form. With Nera, it doesn't. There's no plaintext sitting on a server to breach, no employee, ours or the model provider's, who could see it even if they wanted to, and nothing readable for a third party to compel access to. Privacy stops being a policy decision and becomes a property of the system.
In practice, using it looks like this: encrypt a file, upload it to the chat app, and ask it questions the same way you'd talk to any AI assistant. A few things firms are already using it for: Scanning a trial balance during close and flagging balances that moved more than expected. Calculating employer payroll tax off encrypted gross pay and checking it against the booked accrual. Qualifying R&D tax credit spend, sorting contractor costs, cloud spend, and QSB payroll offsets correctly. Flagging AR aging risk before it becomes a collections problem. Anything involving free-text judgment calls, like reading a reclassification memo, still requires a plaintext step today. We're upfront about that limitation rather than papering over it. The numeric work, the part that touches the most sensitive client data, is where the encryption story is strongest, and it's also where firms spend the most repetitive hours.
The bigger opportunity isn't just faster close cycles, though. Firms that get comfortable with recurring, secure access to client data start using it differently, moving from "we prepared your return" to "here's what we noticed and what you should do about it." That shift, from compliance work to advisory work, is where a lot of firms want to go anyway. Confidentiality has been the blocker. Once the data can move securely and automatically, the advisory conversation gets a lot easier to have.
If your firm has been holding back on AI because of what it would mean for client confidentiality, that's exactly the problem we built this to solve. You can see it working on your own numbers, not a sanitized demo, in about fifteen minutes.
Book a walkthrough at chatapp.nera.systems.