Confidant is the AI workspace your compliance officer would approve of. Client names, account numbers and case details are masked on your machine before anything reaches the AI — and restored in the answer. With an audit log to prove it.
Get startedBuilt in the UK for accountants, solicitors and clinics.
Your staff are already pasting client work into ChatGPT.
The only question is whether the client's details go with it.
Confidant gives your team the AI they want — without the leak you can't afford.
No IT project, no servers to trust. Confidant works like the AI chat your team already knows — with a guardian in between.
Paste an email, a ledger query, a contract clause. Ask Confidant to draft, summarise or explain — just like any AI chat.
Names, addresses, account and case numbers are detected and replaced with tokens on your machine. You see exactly what was masked before anything is sent.
The AI's response comes back with the real details seamlessly restored — and every interaction is logged locally, ready to show a regulator or client.
The power of frontier AI models, wrapped in the discretion your profession demands.
PII, account numbers, NI numbers, case references — detected and masked on-device before every prompt. The AI never sees the real data.
See every masked item highlighted before you send. Add anything the detector missed with one click, or release a false positive.
Add your client names, matter numbers and internal codewords once — Confidant masks them in every prompt, firm-wide.
Your conversations are encrypted and stored on your device — never on our servers. We couldn't read your chats if we wanted to.
Every interaction logged locally with what was masked — exportable as a PDF, plus a ready-made "how we use AI safely" document for clients and insurers.
One-click starters: "summarise this lease", "explain this HMRC notice in plain English", "draft an engagement letter". Useful from minute one.
Upload a contract, return or case file and ask questions about it — sanitised with the same masking before it leaves your machine.
Invite staff, set a firm-wide masking policy that individuals can't weaken, and see usage at a glance.
Powered by leading AI models under zero-data-retention agreements — your prompts are never stored or used for training. UK/EU-hosted option coming.
Built for professions where confidentiality isn't a preference — it's the job.
Use AI on client work without client data ever leaving your practice.
AI drafting power with privilege intact.
Draft letters and notes without patient identifiers leaving the building.
Also built for recruiters, mortgage brokers, estate agents — and anyone whose clients expect discretion.
We're onboarding a small group of founding firms. Leave your email and we'll be in touch personally — founding firms get lifetime preferential pricing.
No spam — we'll be in touch personally to get you set up. Unsubscribe any time.
On your device. Conversation history is encrypted and stored locally — Confidant's servers never hold your chats or documents. The only thing that leaves your machine is the masked version of your prompt.
No. Confidant uses leading AI models under zero-data-retention agreements: prompts are processed to generate the answer and are not stored or used for training. And because masking happens before sending, the models never see real client details anyway.
Out of the box: names, addresses, email addresses, phone numbers, National Insurance numbers, UTRs, VAT and account numbers, dates of birth and case/matter references. You can also add your own always-mask list — client names, internal codewords — and everything is shown to you before sending.
Confidant is built around data minimisation: personal data is pseudonymised on-device before any processing, nothing is retained server-side, and you get an exportable log of every interaction showing what was masked. We provide a DPIA template and a plain-English "how we use AI safely" pack for your clients and insurers.
Simple per-seat monthly pricing, sized for small firms — final pricing is announced at launch. Founding firms who join the early access list get lifetime preferential pricing.
You can — but surveys consistently show staff use AI anyway, on personal accounts, invisibly. A ban you can't enforce is a risk you can't see. Confidant gives your team a better tool than the one you're banning, so the safe path is also the easy path.