CaseFile AI
EN हिन्दी
Login →

One matter, read once.

CaseFile AI is a matter workspace. Each matter has its own documents, its own case file and its own chat. Nothing crosses between matters.

Documents

Upload PDFs and images up to 25 MB. Each document moves through a visible pipeline — uploaded, virus-scanned, text extracted (OCR for scans), pseudonymised, indexed, case file updated — and you see where it is. Failed steps say what failed. Documents that look like investigation material are flagged as sensitive and kept out of drafting. Every document shows how many case-file items it produced and how many you have reviewed.

Case File

Facts, figures, timeline events and open issues, each with its source page. A summary strip shows what is unreviewed, checked, confirmed, corrected, excluded and blocking. Filter by status or document, search the text, page through, or use "Next unreviewed". Open any row for the full quote, the page link, its history and the actions: Confirm, Correct (with reason), Do not use (with reason). Every action is audited.

Case File facts ledger: rows marked AI extracted, one marked Checked against page, and one marked Entered by you — not from a document, each with its document and page.

Integrity checks

Deterministic checks run on every extraction: does the figure match the number in its own quote (lakh and crore understood), does the date appear in the source, do totals reconcile with their components, does the bank statement's running balance add up. Failures become blocking issues with a stated reason — never a silent guess.

Ask

Questions in English or Hinglish, answered from the case file. Each factual sentence carries a pointer to a document page; click it to open the page and see the quote highlighted. Analysis is marked as analysis, not fact. When the record does not contain the answer, the reply says so. History is kept per matter with the same pointers.

Source pages

Every pointer opens the exact page, with the cited text highlighted. Page through the document from there. Each page you open is logged.

Statutes

See the Statutes page.

Drafts

Drafting is not part of the current pilot.

See it

Home board: four stat cards (today's hearings, needs review, blocked, no next date), a seven-day hearing strip, and matter cards with Blocked and Needs review badges.
Every matter's status on one board.
Case File facts ledger: rows marked AI extracted, one marked Checked against page, and one marked Entered by you — not from a document, each with its document and page.
Every fact shows its review status and the page it came from.
Draft legal notice with the Structure pending lawyer review label, a placeholders panel for the advocate, and the first footnoted fact on record (re-cropped to exclude a pseudonymiser leak).
Drafts carry their sources as footnotes and wait for the advocate.
Negotiable Instruments Act, 1881 section 138 shown as official text, amended up to 2018-04-16, reproduced verbatim inside a draft.
Statute text appears verbatim, with the date it was amended up to.

You have tried asking a chatbot about your case. This is why it did not work.

What goes wrong with a general chatbotWhat CaseFile AI does instead
It forgets your documents. You paste pages in, it loses them by the next question, and you start again.Your documents are read once into a case file. Every question is answered from that file, in this matter, with the page each fact came from.
It cannot show you where a fact came from. You get a confident paragraph and no way to check it.Every fact carries a pointer. DOC_1 · p.3 — click it, the page opens with the quote highlighted. If it is not on record, the answer says so.
It writes law from memory. Section numbers and wording that may be old, renumbered or invented — and it cannot tell you which.Statute text is shown verbatim from the official India Code print, labelled text-verified or pending. Never generated. No case-law citations.
Your client's phone, PAN and account numbers go to the AI as they are.Phone, email, PAN, Aadhaar, account and IFSC numbers, UPI IDs, vehicle numbers and GSTIN are tokenised before the text goes to any AI. PHONE_1, ACCT_A. Names and addresses are tokenised only in the forms the rules recognise. Delete the matter and it is gone, with a receipt.

A general chatbot is a good place to think. It is not a place to keep a case.

Who it is for

Advocates with a heavy caseload who want the record read once and reliably. Also used by professionals who draft procedural documents and by litigants preparing their own matter — the same engine, the same rules.

For advocates

The record of each matter, read once and kept — so preparation starts from the case file, not from the folder.

  • Every matter isolated: documents, case file, chat.
  • Facts, figures, dates with source pages; blocking contradictions surfaced before you rely on a figure.
  • Hinglish or English questions; answers that cite the page; statute text with its verification tier.

Limits: no hearing calendar. No case-law citations. Drafting is not part of the current pilot.

For professionals who draft procedural documents

Procedural navigation and document reading, with the same source discipline.

  • Read notices, applications, replies and orders into a case file.
  • Ask what a document says and where; get the section text verbatim.
  • Statutory references you can check against the official print.

Limits by design: no evidence-side coaching, no witness or investigation strategy; investigation material is flagged and excluded from drafting.

For litigants preparing their own matter

The same engine the advocates use, with nothing hidden.

  • Upload your orders, complaint, bank statements; see what the record actually says.
  • Know what is not on record before you rely on it.
  • Prepare questions for your advocate from facts with page numbers.

Limits: this is preparation, not advice. Every draft carries an advocate sign-off block.

How it works

How it works in full: upload, read (OCR), pseudonymise, case file, ask Upload PDF or scanvirus-scanned first Read (OCR) Text is read from each pagescans: read in full by the AIprovider for OCR, before anymasking Pseudonymise applies to the extracted text+91 98… → PHONE_1a/c 4486… → ACCT_Anames and addresses are notmasked Case File Facts, figures, dateseach with DOC_1 · p.3contradictions flagged Ask question → answerwith page pointers"not on record" when absent How it works in full: upload, read (OCR), pseudonymise, case file, ask Upload PDF or scanvirus-scanned first Read (OCR) Text is read from each pagescans: read in full by the AI provider forOCR, before any masking Pseudonymise applies to the extracted text+91 98… → PHONE_1a/c 4486… → ACCT_Anames and addresses are not masked Case File Facts, figures, dateseach with DOC_1 · p.3contradictions flagged Ask question → answerwith page pointers"not on record" when absent

Phone, PAN, Aadhaar and account numbers are masked before the AI reads the text; names only in the forms the rules recognise. A page that cannot be read from its own text, such as a scan, can be read in full by the AI provider for OCR before any masking; masking applies to the text read from it. Every fact keeps its page.

What happens to an answer before you read it

What happens to an answer before you read it: the AI draft passes four checks — page, number, statute, scope — before it is shown AI draft Page check × a sentence with no pagebehind it is not shown Number check × an amount that does notmatch its own quote isflagged as blocking Statute check × section text only fromthe official print, withits label Scope check × no evidence coaching, nocase-law from memory, nooutcome predictions What you read to verify What happens to an answer before you read it: the AI draft passes four checks — page, number, statute, scope — before it is shownAI draftPage check×a sentence with no page behindit is not shownNumber check×an amount that does not matchits own quote is flagged asblockingStatutecheck×section text only from theofficial print, with its labelScope check×no evidence coaching, nocase-law from memory, nooutcome predictionsWhat you readto verify

What passes the checks is what you read. What does not is listed as 'to verify'.

What you can rely on

  • Source-backed Every fact in an answer points to a page in your own documents. If it is not on record, the answer says so.
  • Hosted in India — files in Mumbai, case data in Bangalore Uploaded files are stored in AWS Mumbai; the Case File and chats in Bangalore. Certain AI processing occurs outside India — see Security & data.
  • Advocate in control The software prepares; the advocate decides. Nothing is filed, no outcome is predicted, and every action on the case file is logged.

CaseFile AI is a legal information and preparation tool. It is not a lawyer, does not give legal advice, does not file anything, and does not predict outcomes. Every draft it produces is for an advocate's review and sign-off.

Access is by invitation while the product is in its pilot. What the pilot includes →

Request access