chronolipi
Three dates, always stored
Document date, event date and capture date are separate fields; every view and filter can use any of them.
Features
An archive that reads, sorts and proves
Six capabilities, each doing one job: getting paper out of the rack and into a timeline you can question in your own language — without ever trusting us with the contents.
Core engine
Chronological organisation
Files belong on a timeline, not in a folder tree
Chronolipi reads a date out of the document itself — the deed date, the order date, the report date — and hangs the file on your archive’s timeline. The folder you dropped it in, the filename your clerk invented and the day you scanned it are metadata, not the organising principle. Conflicting dates on one page (a 1998 deed stamped in 2014) are both kept and labelled, so you see the ambiguity instead of trusting a wrong sort.
Auto-series
Related pages, counterfoil pairs and a file’s annexures are detected and bound into one record.
Gaps are visible
The timeline highlights missing years — “no property records for 2003–2007” is a finding, not a shrug.
Any sort, instantly
By plot, by party, by clerk, by urgency — the timeline is the default, never the only view.
Reading
Multilingual OCR
Odia, Hindi and English on the same page
Indian paper mixes scripts mid-sentence: an Odia heading, an English table, a Hindi remark in the margin, a number in Devanagari. Chronolipi flattens and cleans up each photo or scan, then reads it with Tesseract in your own browser, in every language you tick, before anything is encrypted. For handwriting, torn or very old pages you can opt in, page by page, to a sealed AI reading on our server — and the original image is always sealed unchanged next to the text.
Handwriting is honestly harder than print, and Odia cursive more so than Latin print. Our aim is not a global accuracy claim — it is that every uncertain word is marked, reviewable in seconds, and never presented as fact.
14 scripts, no config
Odia, Hindi, English, Bengali, Telugu, Tamil, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Urdu, Assamese, Sanskrit.
Confidence per word
Doubtful words are highlighted for review before sealing, so a wrong survey number never silently becomes your truth. Every number an AI reading produces must also be seen by a second, independent reader — or it is flagged for you to check.
Corrections stay yours
Your fixes are sealed with the document and are never used to train any model.
Numbers that matter first
Dates, amounts, survey numbers, registration numbers and names are extracted as typed fields, not loose text.
Retrieval
AI archive search
Ask a question, get the page
Keyword search needs you to remember the wording of a ten-year-old letter. Chronolipi searches by meaning: ask for “ज़मीन के कागज़”, “jamin” or “land papers” and it finds the sale deed, the patta and the ଖତିୟାନ, in whichever of English, Hindi or Odia they were written. Small spelling slips and word endings are forgiven. All of it runs in your own browser over text only you can unlock, and Ask AI answers from the papers it found, naming each one.
Across three languages
Ideas like land, insurance, tax, school, health and court are matched in English, Hindi, Odia and romanised Hindi.
Forgiving
“deeds” finds “deed”, “ଜମିର” finds “ଜମି”, “insurence” finds “insurance”.
Says how it matched
Each result shows the word that matched, so you can see why it came up.
Private by design
Search runs on your device; the server never sees what your papers say.
Security
Zero-knowledge encryption
Your key never leaves your device
A fresh 256-bit key is generated per file in your browser or app and sealed with a key-encryption key derived from your passphrase with Argon2id. The sealed key travels with the ciphertext. Our servers store, copy, back up and serve bytes they cannot turn into a document, and our code has no path that holds your passphrase long enough to misuse it.
Per-file keys
One compromised key costs you one document, not one archive.
AEAD, not CBC
AES-256-GCM gives confidentiality and integrity together — a flipped bit fails decryption.
RAM-only processing
Decryption happens in an ephemeral container with no network route and no temp file.
Key rotation
Re-wrap every sealed key in one action when a staff member leaves; files never move.
Evidence
Tamper-proof fingerprints
Proof that a paper was not changed
Every document gets a SHA-256 fingerprint the moment it is saved, kept in an activity list that can only grow, never be edited. Each time the document is opened the fingerprint is checked again, so a changed file cannot pass as the original. For court, you print a copy pack: the paper, its fingerprint, who saved and opened it and when, and a draft certificate under Section 63 of the Bharatiya Sakshya Adhiniyam, 2023.
Checked every time it opens
If a stored file no longer matches its fingerprint, it will not open as if nothing happened.
Court copy pack
Fingerprint, times, who did what, and a draft Section 63 certificate (which replaced Section 65B) for your advocate to complete.
An activity list that cannot be edited
Each entry is chained to the one before it, so a removed or changed entry shows.
Auditor access
Give a CA or inspector view-only access to chosen folders and the activity list, until a date you set.
Adaptability
Sector templates
Extraction that knows what a deed is
A loan file, a prescription and a mutation application differ in what counts as the important line. Templates define which fields to pull, which date governs, which parties must be linked and what triggers an alert — “limitation period nears”, “test report missing”, “KYC expired”. New record types are configuration, not a code release.
24 built-in types
Deed, mutation, court order, FIR, discharge summary, prescription, loan application, GST notice, invoice, mark sheet…
Your own types
Define fields, regex and validation rules in the admin panel and backfill existing files.
Rules that interrupt
Expiry windows, missing annexures and inconsistent names raise work items, not reports nobody opens.
Export without a re-scan
Every field is CSV/JSON on request; your data is never held hostage by your archive.
Versus the alternatives
Why not just use Drive plus a scanner?
Because the hard part was never producing the JPEG. It is finding it in 2031, and proving it is the one from 1998.
| Scanner app / Drive | DMS with OCR | Chronolipi | |
|---|---|---|---|
| How files are found | Filename you wrote | Full-text keyword hit list | Date on the document + meaning of your question |
| Who can read stored bytes | Provider staff & any subpoena | Provider staff & any subpoena | Nobody without your passphrase |
| Sorts 40-year-old paper automatically | No | No | Yes |
| Handwritten Odia / Hindi | You look at it | You look at it | Flagged words + opt-in AI read |
| Proves a file was unaltered | No | No | SHA-256 fingerprint + court copy pack |
| Bulk scan-on-demand of boxes | No | No | Yes, on Bulk plan |
| Typical cost per searchable page | Low, but search is unpaid labour | ₹6–12 outsourced | ₹0.40–0.75 at volume |
Implementation detail
Feature questions
Can I upload PDFs I already have, instead of scanning?
Yes — PDFs, TIFFs, phone photos and whole folders via drag-and-drop or the desktop agent. Existing PDFs with a text layer skip OCR for the text and still get date extraction, hashing and indexing.
Do you need an internet connection in the office?
The mobile app queues encrypted files locally and uploads when a connection appears — useful for a tehsil with patchy broadband. A field kit can carry batches on an encrypted drive if needed.
What accuracy should I expect on my paper?
Clean typed pages in English or Odia: most pages need no correction. Old carbon copies, ball-pen overwrites and thermal-faded bills: expect a review queue. We will run 20 sample pages free and show you per-line confidence before you commit to anything.
Is the AI a chatbot that might make things up about my records?
Retrieval is constrained to your permitted documents and every answer carries page citations. It is a search-and-quote system with a language interface, not a source of facts. Where nothing relevant exists, it refuses rather than composes.
Get started
Pick one record type and try it
Tell us the paperwork that hurts most and we will configure the template with you — before any invoice.