How a raw file becomes a record that survives discovery.
The connector is the easy part. The hard part is everything that happens next — to every
email, PDF, photo, statement, telemetry stream, and voice note you let in. Seven stages,
each one inspectable, each one reproducible. The pipeline is what turns a Tesla session, a
14-line HELOC statement, or a parking-lot voice note into a typed record the
Provenance Ledger will sign and keep forever.
A file you can't find isn't a record. It's just clutter.
Most of your life already lives in files — but a PDF in a folder doesn't know what it is,
who it's about, or when it matters. The pipeline is the part of Lossless that turns a pile of
documents into something you can actually ask questions of.
✓ It runs on its own — connect an account and walk away
✓ It reads what people can read: scans, handwriting, audio, video, photos
✓ It files every record into 19 domains and 130+ record types
✓ It writes a plain-language summary for every single record
✓ It never edits or deletes the original — only ever reads
What you connected
scan_2026-05-12_0007.pdf — 2 pages, no text layer, slightly skewed
↓
What the pipeline gave back
Auto insurance — policy renewal
DomainVehicles → Vehicle Insurance
RenewsJun 1, 2026
Premium$1,284 / 6 mo ▲ $90
CategoryVehicle Expenses → Vehicle insurance
Summary writtenSealed · 1 source
The seven stages
Open any stage. See what's actually happening.
Under the hood there are 51 individual steps across 7 stages. You don't need to think about
any of them — but you should be able to see them. Tap a stage to look inside.
Nothing sits in a folder waiting to be noticed. The instant a file comes in — from your inbox,
your drive, an upload, a voice note — Lossless gives it a stable address and opens a record for it.
From here on, it can't get lost.
Given a permanent, addressable home
A record of its own is opened
A scanned receipt isn't text until someone reads it. Lossless reads everything: the words on a page,
the words inside a photo, what's spoken in an audio note, what happens in a video. If it's in another language,
it's translated. It even reads the quiet details — when a photo was taken, where, and on what device — and
picks out the people, businesses, amounts, orders, and dates along the way.
Printed text & handwriting (OCR)
Images inside documents
Audio transcribed
Video transcribed & narrated
Photos described
Anything translated to English
When & where it was created
The file's own metadata
People mentioned
Businesses & organizations
Amounts & money
Orders & receipts
Events & dates
Anything you've taught it to look for
Now the file is understood, not just transcribed — what it's about, who it concerns, why it matters,
and what (if anything) you need to do about it. Lossless quietly connects the new file to the ones it
belongs with: the confirmation email, the earlier statement, the right person, the right property.
The pile starts becoming a story.
What it's about
Key quotes pulled out
Action items spotted
Routed to the right handler
Linked to related files
Tied to the right people
Attributed to the right property
Patterns & behavior noticed
Importance scored
Sorted into clean, structured fields
Every record gets a short, plain-language summary and a one-line brief — the kind of thing you'd
jot on a sticky note. It's categorized — filed into the taxonomy of domains, document
families, and spend categories. And if something's missing or thin, Lossless goes back and fills
the gap rather than leaving a hole.
Filed into a category
Classified by document type
Gaps healed
A plain-language summary
A one-line brief
How relevant it is to you
A record is only useful if it shows up when you need it. Lossless tags each one with the topics and the
bigger themes it belongs to — drawn from a three-tier map of your life that the system builds for you —
so a question three years from now still finds the answer.
Topics extracted
Themes assigned
This is the part the name is about. The finished record is kept in several places at once, made searchable
by meaning rather than just keywords, woven into the graph of people and places and things you own,
and sealed with its provenance — payload, origin, seal. The original file is never touched.
Nothing gets lost.
Searchable by meaning
Stored in the database
Original kept in cloud storage
Added to the vector store
Woven into your entity graph
Metadata written back
Sealed with provenance
Before Lossless calls a record done, one more pass looks it over — checking for anything thin, missing,
or off — and fixes it. The bar is simple: would this hold up if you actually had to rely on it?
A final quality review & gap repair
7 STAGES · 51 STEPS · RUNS IN THE BACKGROUND · YOU NEVER PRESS A BUTTON
Inside the app
You can watch it work.
Open any record — or any upload batch — and there's an AI Pipeline tab.
Every step, timed. Every file in a batch, mapped. Nothing hidden behind a spinner.
lossless · records · evidence_photo_001.jpg
Live
The Overview tab shows the AI summary, the one-line brief, and every extracted entity —
all six of which are covered further down this page.
68%
Elapsed
0:41
Remaining
0:19
Steps
35/51
Stage
3 / 7
Currently running: Topic Extraction — AI Analysis
📥Ingestion2/2
🔍Extraction16/16
🧠AI Analysis7/15
✓Scores Extraction1.4s
✓Quotes Extraction2.1s
✓Action Items Extraction1.8s
✓Document Routing0.9s
✓Related Files Discovery3.2s
✓Person-Document Linking2.0s
✓Behavior Pattern Analysis2.7s
Topic Extractionrunning…
·AI Structured Extraction—
✨Enrichment & Summarization0/6
🔗AI Search Readiness0/2
💾Persistence0/9
✅Quality Assurance0/1
186
Complete
49
Processing
0
Pending
5
Needs review
File completeness heatmap — 240 files in this batch
Complete Processing Pending Needs review
Every cell is one file. Hover to inspect. A batch of years of paperwork runs the whole
51-step pipeline, file by file — and tells you exactly which five need a human glance.
The Entities tab lists every person, business, place, amount, date, topic, and action item the
pipeline pulled — each one tappable, each one linked back to the source.
Follow one file
Pick something ordinary. Watch it go through.
The pipeline treats a voice memo and a nine-page bank statement with the same care. Choose one and see.
Watch the pipeline handle
an insurance renewal PDF
a voice memo from the car
a bank statement PDF
a photo of a receipt
The taxonomy
Every record gets a place.
"Categorized" isn't a vague promise here. There's a real, structured taxonomy underneath —
domains, record types, document families, and a deep spend map — and every file is routed into it.
Step 1
A file arrives
Any type — email, PDF, photo, statement, voice note.
"scan_0007.pdf"
Step 2
Routed to a domain
One of 19 life domains — the broad area it belongs to.
→ Vehicles
Step 3
Typed as a record
One of 130+ record types — what kind of thing it is.
→ Vehicle Insurance
Step 4
Filed by category
Document family + spend category, down to the sub-category.
→ Vehicle Expenses → Vehicle insurance
Step 5
Tagged & linked
Entities, topics and themes — so it surfaces on demand.
→ people · amounts · "Auto & Insurance"
19 domains — the broad areas of a life
Every record belongs to one. Together they hold 130+ record types and 25+ sub-types.
DentalPharmacySupplementsVisionMedical insuranceMedical appointmentMedical procedureMedical suppliesLabworkSkin treatmentsAddiction treatmentTherapySpaMassageYoga classGym membershipPersonal training
+ Groceries & Food · Housing · Digital Services · Shopping & Retail · Travel · Transit · Education · Childcare · Charity & Gifts · Professional Services · Beauty & Personal Care — 21 in all
What you get back
Six things every record carries — that the raw file never did.
The plain-language summary
A short, human paragraph that explains what the record is — readable in about five seconds, no jargon, no skimming a PDF.
The one-line brief
An entity-first headline with the significant stuff flagged — the deadline, the amount, the thing you'd actually want to know first.
The right shelf
Routed into the taxonomy — a domain, a record type, a document family, and a spend category down to the sub-category.
The entity graph
Every person, business, place, amount, and date it touched — linked, so you can walk from a person to a property to an account.
Themes & topics
The throughlines of your life, tagged onto the record across a three-tier topic map — so a question years later still surfaces it.
The provenance seal
Payload, origin, seal. Every finished record carries proof of where it came from — so any agent you authorize can trust it.
By the numbers
One pipeline. Every kind of file you own.
7stages · 51 steps
19domains · 130+ record types
39document families
100+spend sub-categories
"I connected eighteen years of Gmail and a folder of scanned paperwork I'd been avoiding.
A day later it was all just… records — summarized, dated, filed under the right category.
I didn't lift a finger. The pipeline did the part I'd been dreading for a decade."
"Reproducibility at the parser level is the part the security auditor actually cares about — re-run the source, re-derive the record, the signature still verifies."
— What the technical evaluator writes in their report
See the pipeline doing real work.
Connect one account. Watch a few years of statements, receipts, and telemetry turn into typed records the ledger will sign. Reads, never writes.
Next pillar: the schema-aware connectors that hand the pipeline its raw material — Gmail receipts, Plaid statements, Tesla telemetry, iMessage threads — each one parsed into a typed record, not a blob.
