01 · Buyer
One buyer, not a team account.
Agency or solo, years deep in one niche, paying from their own pocket because the relationships belong to them.
Meet the buyer →Recruiters re-source people they already know because what they know is not retrievable when it matters.
02 · Problem
They re-source people they know.
The right context is trapped across memory and six tools, so a live brief sends them back out to strangers.
See the pain →03 · Bet
Soft context is the asset.
A CV records what someone has done. The placement turns on what they want, what they said, and when it was true.
Test the bet →04 · Solution
A private book that travels.
Every useful memory is findable, dated and sourced—and the recruiter can export it whenever they choose.
Define the product →05 · Product logic
Two moments. Nothing else.
Capture when something is said. Retrieve from a brief or a name when the moment comes.
Watch it work →06 · V1 proof
Prove a habit and a payment.
Ship complete memory workflows, then ask two hard questions: is capture alive in week six, and will a stranger pay?
See version one →07 · Open decisions
Honesty before we build.
Measure missed-brief impact, set the extraction-cost rule, and choose the first-wave extraction scope.
Open the questions →Why now
- Building cost collapsed
- The niche stayed valuable
North star
- The book outlives every job
- Never re-source known people
Later
- Voice and filter views
- Automate after capture habit
Never
- No ATS imports or scraping
- No bulk, teams or AI ranking
01 · Start here
Your own candidate database.
It remembers what you know, not just what’s on the CV — and gives it back the moment you need it.
The whole case in two minutes: the buyer, the bet, the product, version one and the boundaries that protect it.
The product case2:00
What We’re Building
A direct builder-to-builder account of who this is for, what version one must prove and what recmory refuses to become.
Read transcript →02 · Who this is for
The recruiter who owns the relationship.
A recruiter who pays for it themselves. Years in one niche, thousands of relationships they built with their own hands. Agency or solo — it does not matter. The relationships are theirs.
The screening test
They can name three people right now they know they know, and cannot find.
Job title does not predict payment. Ownership economics do.
They have no patience for a complicated CRM. They need a private book that travels with them and becomes more valuable instead of resetting with every job.
03 · The pain
They cannot get back to what they already know.
The knowledge exists — in their head, in six apps, in a notebook. It just is not retrievable at the moment it is worth money.
Functional
Find the right person from imperfect memory, see what happened before, and know what has gone stale.
Social
Be the recruiter who knows people, not the one who searches for people.
Emotional
Feel prepared and credible — rather than ashamed of forgotten context and a messy spreadsheet.
Door A hurts most. Door B is not optional.
An unfilled brief is an unearned fee. But the root cause is capture: “wife’s pregnant, won’t relocate” has no field, so it stays in the recruiter’s head — and their head forgets.
04 · What they do today
Tuesday-morning archaeology.
A brief lands. One name surfaces from memory. Then the hunt moves through four places that cannot answer the whole question.
The pain, before the product1:04
The Three You Can’t Find
A brief arrives and Leila already knows three people who might fit. Then the tools she has turn fifteen years of relationships into another search for strangers.
Read transcript →Their head
“Someone at Acme. We spoke in 2022. Wanted Berlin. Not ready then.”
one fragmentThe current identity is there. The conversation is not. The search returns 400 results.
400 resultsThe spreadsheet
Thirty rows say “Germany.” None says why this person might move now.
30 × GermanyA scroll through old threads for a conversation they half-remember. Then the next fire starts.
context missingSo they source three new candidates from scratch, because it’s faster than finding the ones they already know.
05 · The bet
A person is not their CV.
Every tool stores what a candidate has. What wins a placement is what they want.
What they have
The CV
What they want
The person
wants a promotionwon’t move while the baby is smallwould leave for the right board seatburned by their last boss
The Resume is only the commodity. It is the soft context that is the asset.
06 · What recmory is
Capture. When the thing is said.Retrieve. When the moment comes.Intelligence. Surface the context; sharpen recruiter judgement.
The positioning
Your own candidate database.
For a recruiter who buys their own tools — years in one niche, thousands of relationships they built themselves — that needs to get back to what they already know about someone, at the moment it is worth money, recmory is your own candidate database that remembers what they know, not only what is on the CV, and gives it back the moment they need it.
Unlike the spreadsheet, the CV folder and their own memory, recmory provides one place where what a candidate said is as findable as what is on their CV — with a date attached, so they know what still to trust.
The honest competitor
Not Bullhorn. “Good enough, already open.”
The spreadsheet. LinkedIn. Their own head. This buyer is not shopping for a smaller agency ATS; they are trying to recover context from the tools already in front of them.
Two doors. One store.
The same memory answers from opposite directions.
07 · See it work
From the first file to the right call.
Three films. Three working proofs. Everything interactive runs locally against fictional data; nothing is uploaded or saved beyond this page.
01
The first minute.
Start with the mess the recruiter already has. Recmory works out the columns, names what it cannot understand, and asks the user to inspect only the uncertain rows.
fictional file and import reportImport without reformatting1:00
The First Minute
A recruiter drops in the spreadsheet they already use. Recmory reads the mess, protects the uncertainty and gives them a useful first result before asking them to learn anything new.
Read transcript →MY CANDIDATES final FINAL 4.xlsxSheet1 (2431)
| Full name | Curr. Co | Notes/comments | ? |
|---|---|---|---|
| Voss, Elara | Northstar Gds | baby small / spring? | blue |
| Jon Bell | Helian | 165% metric on old CV | |
| AMINA RAHAL | — | via Marta; board seat | 4 |
Full name→name
Curr. Co→company
Notes/comments→notes
?→I can’t tell — review
2,431 people imported. 12 rows need your eye.
You never reformatted the file. Unknown column “?” was kept aside rather than guessed.
02
The twenty-second note.
The whole bet rests on whether a useful human detail can be kept before the next call starts.
fictional person and noteThe capture habit1:08
Twenty Seconds
A nineteen-second note in a car park sits untouched for eighteen months. Then one word brings it back at exactly the right moment.
Read transcript →Quick note
Elara VossFinance Director · fictional
recorded 4 Aug 2026as of Jul 2026
03
Two doors. One memory.
A brief finds the person. A name retrieves the context. The same store answers from opposite directions—and shows why every answer matched.
fictional brief, people and historyRetrieval in both directions1:20
Two Doors
A live brief finds Mara through a sentence no CV contains. A name opens Devon’s history and shows why calling him would be the wrong move.
Read transcript →A
A brief lands.
Search the words you remember. Try “won’t relocate” to find something no CV knows. Every result explains itself.
The note beat the CV.This person exists in the result because of Leila’s own words. The phrase appears nowhere on the CV.
1 resultranked by matching evidence
B
A name appears.
Open a person and retrieve the whole story: sourced facts, dated memory and every touchpoint. Staleness is shown, not hidden.
Elara Voss
Finance Director · Northstar Goods
Known facts
Stale — verify before usingMunich · open to relocate
You wrote this 3 years ago.
08 · What ships first
One system.Seven obligations.
Capture feeds retrieval. Retrieval is useless without trust. Trust is useless if the first import fails, the data is not theirs, or nobody can buy the product.
A
Capture
A1
Add a person fast — name plus one way to reach them, everything else optional
SA2
A note against a person in ~20 seconds, from a phone
SA4
as_of on facts — when it was true, not only when it was written
B
Retrieval · door A
B1
One search box across name, note text, city, company and title
MB2
Search by phone number
SB3
Results show why they matched — the snippet that hit
MB7
Search finds text inside CVs — the “165%” case
MC
Retrieval · door B
C1
Person page — everything known, dated
MC2
Timeline of every touchpoint
MC3
One person across roles — today’s candidate is tomorrow’s client
MD
Trust · can I believe this?
D1
Every fact shows where it came from and who entered it
SD2
Stale facts flagged loudly — “you wrote this 3 years ago”
SE
Getting started
E1
CSV / XLSX upload where the system works out the columns; no restructuring
LE5
Upload a CV (pdf/docx) and extract fields into the person’s profile
LE6
The user confirms or corrects extracted data before it is saved
ME7
The whole CV text is stored even when nothing was extracted
SF
It is yours
F1
Full export, readable format, one click
SF2
Delete everything, for real
SF3
Plain English: where the data lives and who can see it
SG
The business — without these nobody can pay
G1
A landing page that says who this is for
MG2
Sign up and log in
MG3
Take money — Stripe
MG4
Privacy policy
SG5
Terms of service
SG6
A visible price
SG7
Empty states everywhere
MG8
The import failure report
SG9
A way to reach a human
SG10
Log out, and change email
SG5–G10 added 5 August 2026. Group G was in v1 and appeared in no workflow, while v1’s own bar is a stranger who found recmory online pays for it. A stranger cannot pay a price they cannot see, and cannot lawfully be charged without terms.
S = days · M = 1–2 weeks · L = 3 weeks or more
09 · How it gets built
Six workflows.One launch.
Revised 5 August 2026. The first list of five was derived from the feature menu and approved as a table, not decided. Under examination it did not hold: one PRD was two jobs, one was a bundle with no trigger, one feature sat behind the wrong door — and the entire launch group, which v1’s own bar depends on, appeared in no workflow at all.
Get my history in
“My fifteen years are in here and I did not retype them.”the spreadsheet — unlocks every retrieval flow
Get my profiles and CVs in
“The documents I already had are now people I can search.”plus the minimum duplicate matching — v1, not later
A job opportunity arrives, find the candidate
“A brief arrives and I find the people I already know.”depends on imported searchable history
A name comes up, retrieve the insights
“A name appears and I know within seconds when we last spoke and what I promised.”the same store, the opposite direction
Capture new insights
“I said something worth keeping and it took twenty seconds to keep it.”two entry paths, one PRD — a new person, or something new about an old one
Leave with everything
“I can take all of it with me, and delete it for real.”makes the private book credible
Launch-oriented · not a workflow in the same sense, and kept apart so it cannot be quietly dropped
A stranger arrives and pays
“I found this on my own, understood it, and paid for it without speaking to anyone.”the only PRD v1’s own bar directly depends on
The cross-cutting checklist
Not a workflow. A list every PRD above must answer, so it is never forgotten in one of them.
- empty state · loading state · error state
- provenance shown — D1 is a rule inside PRDs 1, 2, 4 and 5, never a PRD of its own
- works on a phone
- what happens when the connection drops
Before PRD 2 can be finished
Set the CV-extraction cost rule — 5,000 CVs at $0.10 is $500 to onboard one user — and decide whether CV extraction belongs in the first shipped wave at all. Both land on PRD 2.
10 · Later, and never
Scope is what the product refuses.
Real. Wanted. Not now.
Later
Roughly in the order these are expected to become urgent.
Full duplicate detection and merge, with undoAmended 5 Aug 2026. The minimum — matching an incoming CV to a person already there — is v1, inside PRD 2, because PRD 1 and PRD 2 import the same people and the two collide on day one. Everything beyond that minimum still waits.
One next action per personCapture has to be a habit first.
Voice note → textThe obvious upgrade once the note proves itself.
Filters and saved shortlistsThey matter when search returns too much.
Who have I not spoken to in twelve monthsIt needs enough history before it is true.
LinkedIn connections export importA second import path; the spreadsheet goes first.
Quick capture outside the appWhere capture wants to end up.
Phone rings → context appearsThe best idea in the research and the hardest to build.
Auto-log email and WhatsAppThe Recruiter-OS end state.
Unverified and source-unknown shown loudlyCheap, simply not load-bearing yet.
Parse “last month” into
as_ofThe field ships now; only the parser waits.Refused, on purpose.
Never
This is the more important list. It protects the reason the user trusts recmory.
No employer ATS or confidential client-list importsThe relationship travels; the record does not.
No LinkedIn scraping or prohibited automationOne cease-and-desist ends the company.
No bulk email sequencing or mass outreachThe recruiter is the warm channel. Automation destroys that asset.
No job posting, timesheets, compliance or client portalsThat is an agency ATS: different buyer, different company.
No team workspace or shared system of record“My system, not my employer’s” is the ownership promise. Team revenue is the most tempting drift on this page.
No AI candidate ranking or scoringThe recruiter’s judgement is the product. Ranking it insults the user and adds liability nobody asked for.
No third-party scripts touching a real person’s dataNames, email, phone and notes stay away from analytics and session replay.
11 · Where this ends up
The recruiter’s operating system.
The first thing they open and the last thing they close — because the thing their day is actually made of, people they know, finally lives somewhere that answers.
The book outlives the job.
A recruiter changes agency four times in ten years and loses nothing. Their memory compounds instead of resetting.
Nobody re-sources someone they already know.
The first place they look is their own memory, because it finally answers.
Memory is what they sell.
Clients pay for reach into people the recruiter already has trust with — not for searching LinkedIn faster.
Recruiter OS is the destination. Capture and retrieval is the road.
12 · Why now
The market was always there. It was never worth building for — until building got cheap.
Nobody bothered
Too small for venture money, too niche for ATS players. Everyone sold top-down to agencies.
Building got cheap
What needed a team of ten now needs two people and AI.
Underserved, not unvalued
The value was never in question; the old cost of reaching it was.
The ramp is short now
A small team can reach a usable product in months, not years.
13 · What we don’t know yet
No smoothing over the gaps.
If something is unknown, it stays visible until evidence settles it.
What does one missed brief cost?
The pain is clear. Its commercial impact is not. This comes from direct problem interviews, not arithmetic made up on this page.
What is the CV-extraction cost rule?
5,000 CVs × $0.10 = $500 to onboard one user. Extract everything, extract on first open, or cap the first batch? A cheaper model does not replace the product decision.
Does CV extraction belong in the first shipped wave?
Searchable CV text serves door A. Full extraction adds another variable to the first test. The PM has to choose knowingly.
The evidence weakness
No recruiter has said any of this to our face yet.
The research is practitioner writing plus judgement. The next gate is direct problem interviews and observed retrieval tests.
The risk that kills it
People say they want to capture. Then they don’t.
Every personal CRM in history died here. If meaningful capture is not still happening in week six, stop adding fields. The memory loop has failed.