A private AI admin assistant for law firms and legal teams
For the administration around legal work — fees, interest, letters, checklists — where confidentiality is not a preference but a duty.
Short answer
OpenClerq handles the administrative side of a legal practice without sending privileged material anywhere. Fee caps, disbursements and simple interest run as formulas in a deterministic engine with a proof record; a local model drafts routine correspondence and explains figures in plain language; recurring reminders run on a schedule; every run is recorded. It does not give legal advice, and it is not meant to.
Why legal teams hold back from AI
- Privilege and confidentiality. Uploading client material to a third-party service needs careful thought and, often, client consent.
- Hallucination. A model that invents a figure will also invent a reference. Anything it produces is a draft.
- Accountability. Fee and interest figures sent to a client or a court must be reproducible.
OpenClerq's answer is to keep the model on your machine and away from the numbers, and to keep it on administrative work.
What OpenClerq does for a legal team
| Task | How | Status |
|---|---|---|
| Fees against a cap, with VAT | Formula spec with a proof record | Today |
| Simple interest on a sum over a period | Formula spec, using the rate and day count you specify | Today |
| Drafting routine letters and file notes | A local model, from the template and facts you give it; a lawyer reviews every draft | Today |
| Explaining a fee breakdown in plain language | The model explains the engine's figures for the client letter | Today |
| Recurring reminders | Scheduled triggers for billing runs, file reviews and renewals you list | Today |
| Reading matter files from an allowed folder | The tool-calling loop, inside the policy engine and sandbox | 0.6 |
| Your fee rules, rates and letter templates, maintained for you | A module built for your practice | Module |
Example formulas, checked by the engine
An hourly fee capped at an agreed amount, with 20% VAT on top:
{
"spec": {
"formulas": {
"fee": "min(hours * rate, cap)",
"total": "round(fee * (1 + vat_rate) * 100) / 100"
},
"output_names": ["fee", "total"]
},
"inputs": { "hours": 14.2, "rate": 240, "cap": 3000, "vat_rate": 0.2 }
}14.2 hours at 240 would be 3,408; the cap brings the fee to 3,000, and the total with VAT is 3,600.00. The model can then explain in the client letter that the cap applied and by how much.
Simple interest over a number of days: round(principal * annual_rate * days / 365 * 100) / 100. On 25,000 at 8% for 45 days, the engine returns 246.58. Statutory rates, day-count conventions and compounding differ by jurisdiction and by contract — the formula states which one you used, so anyone can check it.
What stays with lawyers
- All legal advice and every judgement on a matter. OpenClerq's default instructions treat legal advice as out of scope.
- Reviewing and signing every letter a model drafts.
- Checking any reference, citation or date — models invent them.
- Deciding what may be processed at all, under your professional rules.
Privileged material stays on your machine
With a local model through Ollama or LM Studio, prompts and documents never leave the computer — there is no provider to assess, no retention period and no transfer. The gateway answers only requests from that machine carrying its token, OpenClerq's files are readable only by the user running it, and the run history records every model call. If your firm allows a cloud model for some work, the context preview shows exactly what would be sent first. See how to use AI with confidential client data.
Frequently asked questions
Can OpenClerq give legal advice?
No. It is built for administrative work — calculations, routine drafting, reminders — and its default instructions treat legal advice as out of scope. Everything it drafts is for a lawyer to review.
Is it safe to use AI with privileged documents?
Sending them to a cloud service raises questions of confidentiality, consent and retention that your professional rules govern. A model running on your own computer avoids sending them anywhere. Check your regulator's guidance either way.
Can it calculate statutory interest?
It evaluates whatever formula you give it, exactly. It does not know your jurisdiction's statutory rates or conventions; you supply them, or we build and maintain them as a module.
Does it read PDFs and scanned documents?
Not in release 0.5, which reads plain-text files only. For now, paste the text you need into the Ask panel.