Every firm is sealed off.
Every query for firm data is limited to the firm in code, and a link to another firm's record finds nothing, as if it did not exist. Tests check it on every change.
GET /matters/2231404 · not found
Security
Your files stay in your firm.
Each firm is sealed off in code; public court data is the only thing shared. Machines propose, people approve, and you can take everything with you.
Six principles
Built in, not bolted on.
Machines propose. People approve.
Deadlines, tasks and client messages arrive as proposals. Nothing reaches a court or an invoice until someone accepts it. Routine date alerts to clients go out on their own only if the firm switches that on.
Deadline: reply in 30 daysAccept
Consent before WhatsApp.
Client messaging needs consent on file, with an off switch per client and a record of every message sent (DPDP Act).
Consent on file
AI shows its receipts.
AI runs through one gateway: outside text is sealed before the model reads it, answers cite the page or paragraph they rely on, and code checks the quotes; what it cannot verify is marked.
p. 14¶ 22(2019) 4 SCC 197
Backed up, and checked.
The database and uploads are backed up every day, and a weekly restore drill checks the latest backup; the health page shows when each last ran.
Backup verified
Leave with everything.
The firm owner can download all of the firm's data in one ZIP. A client's data is copied or erased on request, keeping only what the law requires. The firm stays in control.
ExportDelete on request
Where the AI runs.
- AI features use Google's Gemini API (generativelanguage.googleapis.com), called only from our server through one gateway.
- On the paid Gemini API, Google does not use prompts or answers to improve its products. It logs them for a limited time only to detect abuse.
- Google may process and cache the data in any country where it has facilities. There is no India-only region on this API.
- When web search is switched on for a question, Google keeps that question and its answer for 30 days.
- On our side, each AI call leaves a ledger line (task, cost and a fingerprint of the input), not the papers. Answers are kept where the firm saves them, such as a draft or a summary, and some for up to 24 hours in a cache so the same question is not paid for twice.
- A client can say no: switch off "AI may read this client's papers" on the client's page, or the client can do it in the portal. AI then never reads that client's matters.
No badges here until an auditor signs one. These are the practices the code follows today, and the tests that hold it to them.