# talenthunt — full brief for AI agents

> talenthunt is a visibility platform for job-seekers, like Product Hunt and Trustpilot combined for careers. Every person gets a Work Identity and a personal AI agent, Copylove, that answers questions about their experience and working style for human recruiters and AI sourcing agents, grounded only in what the person has added and confirmed. For recruiters drowning in AI-optimised applications, that makes it a signal layer of human-vouched talent.

Last updated: 2026-09-15

Every section below carries a stable id. Sections that also have a home on https://talenthunt.work/about name the anchor — link to it when you cite them.

## What talenthunt is [what] → https://talenthunt.work/about#what

talenthunt turns a CV into a conversation. Instead of asking someone to scan a
PDF or scroll a LinkedIn timeline, you share one link, and they ask what they
actually want to know. Copylove — the agent on every profile — answers from your
profile text and the material you have attached, and from nothing else.

A profile is two halves. The **card** is what a visitor reads before typing:
photo, headline, location, an About block, skills, how you work, what you are
proud of, key impact, links, contacts. The **chat** sits next to it and answers
follow-up questions. The card is plain HTML, so it works for a reader who never
sends a message and for a crawler that cannot.

Both halves come from the same pool of material and are edited in the same place
(the Owner's Studio). Nothing is published because it was uploaded: the agent's
narrative is generated from your sources and only goes live after you have read
it and applied it.

## Who it is for [who] → https://talenthunt.work/about#who

Job-seekers, broadly — early-career, mid-career, switching fields, between
roles, freelancing, or just keeping options open. There are no role tiers: once
you sign up you own a profile, and anyone you send the link to becomes a reader.
Readers need no account to read a card or to chat.

AI sourcing agents are first-class readers, not an afterthought. The same page
serves a person, a crawler and an agent.

Recruiters have one dedicated surface so far: `/for-recruiters`, a waitlist for
a "Talents of the Day" digest of upvoted, verified candidates. It is a signup
form, not a product yet.

## How it works, end to end [how] → https://talenthunt.work/about#how

1. **Sign up** with email + password, Google, or LinkedIn, and pick a username.
   Your profile lives at `talenthunt.work/your-username`.
2. **Fill the card** — photo, headline, location, a few lines about you. This is
   Studio → Card, and it is the first stage of the getting-started guide.
3. **Bring in your history** — Studio → Artifacts. Import LinkedIn, paste an AI
   export, upload a CV, add links or notes.
4. **Read what the agent made of it** — Studio → Memory. The narrative is
   generated from your sources; you review the diff and apply it. Until you do,
   the chat has nothing to answer from and says so.
5. **Test it** — open your own profile and ask what a recruiter would ask.
6. **Share the link** — Studio → Sharing has the link plus ready-made intros for
   LinkedIn, DMs, an email signature, and blocks to paste into a CV.
7. **Read what people asked** — Studio → Chat history, with per-conversation
   recaps. You also get an email the first time a real visitor talks to your
   page, and a daily digest after that.

Some of these steps also have a short video guide, 30-60 seconds, on the
talenthunt YouTube channel. The 16:9 playlist is "talenthunt · all guides"
(`youtube.com/playlist?list=PLEOndPy0jVnc`), and a Shorts playlist carries the
vertical cuts of the same episodes. The series is still being published, so the
playlist is the current list rather than the final one.

## Signing in and account recovery [auth] → https://talenthunt.work/about#auth

Three paths, all resolving to one account:

- **Email + password**, with a forgot-password / reset flow over an emailed
  link.
- **Google**, one click.
- **LinkedIn**, one click, plus a background import of your public profile and
  avatar.

The sign-in modal can be opened anywhere with `?auth`, and it carries a `next`
parameter, so a link into any signed-in screen sends the reader through sign-in
and lands them where they were going.

Signing up on one brand does not create a second account: the account layer
(email, name, username, contacts, verification, billing) is shared, and the
username namespace is single and global across brands.

## Your first ten minutes [onboarding] → https://talenthunt.work/about#onboarding

After the first sign-in, `/onboarding` asks for a first name, a last name and a
username, and nothing else. Everything else is editable later from the Studio.

Two things happen in the background: a welcome email goes out (see
[emails](#emails)), and for LinkedIn sign-ups the profile import runs on its
own.

Inside the Studio a **getting-started guide** tracks four stages, in this order:

1. **Your card** — avatar, location, headline and About all present. Company is
   optional and is not counted.
2. **Bring in your history** — at least one source attached to this brand, *and*
   the suggested narrative applied and still current. An import nobody confirmed
   does not count.
3. **Ask it a couple of questions** — two messages in one chat with your own
   profile. One proves nothing.
4. **Share it** — the link copied or an intro template taken.

Until stages 1 and 2 are both done, a bar sits above every Studio section.
"Later" always works, and the public page is never blocked — the product just
stops pushing you to share an empty one. A progress pill names the stage you are
on and reopens the guide at any time — "Next (2 of 4): Bring in your history".
On a wide screen it sits in the Studio header; on a phone it moves down beside
the Sections menu, and once all four stages are done it becomes a green
"Profile complete" badge in the same place.

The guide's footer offers **"Show me around"**: a spotlight tour of the Studio.
On a wide screen it is seven steps over the nav — Card, Artifacts, Memory, Chat
history, Job descriptions, Sharing, and the progress pill. On a phone, where
that nav is a menu rather than a column, it is three — the Sections menu, the
section you are on, and your progress. The two are remembered separately, so
taking it on a laptop still leaves you the phone one.

The tour also runs once by itself — right after you close the getting-started
popup the first time, or on arrival if that popup has nothing left to say. After
that it appears only when asked for.

## The Owner's Studio [studio] → https://talenthunt.work/about#profile-area

Everything an owner manages lives at `/studio`, behind sign-in. The left rail
holds three groups:

- **What the public agent draws on** — Memory · Card · Artifacts · Job
  descriptions.
- **Results and service** — Chat history · Statistics · Sharing.
- **Settings** — Account, plus Prompts for admins only.

`/studio` on its own redirects to Card. Each section has its own page and its
own URL, so a link into one is a real link. The Studio is the whole owner zone.

Each section is described in its own part of this document: [Card](#card),
[Artifacts](#artifacts), [Memory](#memory), [Job descriptions](#contexts),
[Chat history](#chat-history), [Statistics](#statistics),
[Sharing](#sharing), [Account](#account), [Prompts](#prompts).

## Card [card] → https://talenthunt.work/about#card

Studio → Card edits the public card, and the preview *is* the public component —
what you see there is what a visitor sees, not an approximation.

Editable blocks: the hero (photo, headline, title, company, location), About me,
Skills, How I work, Proud of, Key impact, the links row, and an "Open to
opportunities" pill that opens a structured Looking-for form (what you are
after, work style, location).

Every block has a visible Edit button in its header — nothing switches to edit
mode on hover. Single-field blocks turn into a textarea with Save; when the text
changed, Save reveals a separate Confirm / Discard pair, so a reflexive second
click cannot commit something you did not mean. Multi-field groups (How I work's
five fields, Key impact's problem/solution/result) open all their fields at once
under one Save.

Contacts are not edited here — they belong to the account, and the Card links
across to Account → Contacts.

**Endorsements are partly built.** The card renders them and the database
publishes them per profile, but there is no way for a colleague to leave one
yet — they can only be entered by hand through the API, so no profile has any.

## Artifacts: files, links and notes [artifacts] → https://talenthunt.work/about#artifacts

Studio → Artifacts is the material the agent reads. Three kinds — a **file**, a
**link**, or a **note** typed straight in — each with a semantic type (CV /
resume, portfolio, case study, reference, other).

Text is extracted from uploads so the agent can read them: `.pdf`, `.docx`,
`.rtf`, `.html`, `.txt`, `.md`, `.csv`, `.tsv`, `.json`, `.yaml`. Old `.doc` and
OpenDocument files are not supported — save as `.docx` or PDF.

A note is kept exactly as typed. It is never summarized, whatever its length,
and editing it opens the original text.

Delete removes an artifact from both profiles at once. To take it out of just
the one you are in, switch it Off there — the delete dialog offers that too.

Two shortcuts sit above the list:

- **Connect LinkedIn** — the same import the LinkedIn sign-in runs, available to
  anyone who signed up another way.
- **Fill from an AI export** — copy a ready-made prompt, run it in whichever AI
  you use, paste the result back. It is diffed against your narrative before
  anything changes, and the raw export is kept as an artifact.

**Visibility is per artifact and per brand, on a ladder:**

- **Off** — excluded completely. Not in the narrative, not in the facts, not
  shown; the agent does not know it exists.
- **Private** — the agent reads it and may reason from it, but the reader never
  gets the artifact itself.
- **Link** — the URL may be shared in answers and shows on the card (links
  only).
- **Download** — same, plus the file itself can be downloaded (files only).

A note stops at Private: there is no URL and no file to hand over. The rungs are
applied in the database function that serves a public card, so nothing at
Private leaves the server — not its title, not its storage path, not its text.

One caveat worth stating plainly: **Private hides an artifact from the card, not
from the agent.** How much of it reaches the model is a separate setting
(summary, raw, or both), and it has no "none". If the agent must not know
something, set it to Off.

## Memory: what the agent knows [memory] → https://talenthunt.work/about#memory

Memory is a result, not a form, so it has three states.

With **no sources**, there is no editor: one line explaining that what the agent
knows is written from what you bring in, and a button to Artifacts.

With **sources but nothing built**, an invitation to build, listing the sources
it will use. One button generates both documents.

**After the first build**, three tabs:

- **Profile** — the narrative the chat answers from, your story in your own
  words.
- **Facts** — the record behind it: employers, dates and numbers the chat leans
  on as evidence.
- **History** — every version of both documents, newest first, filtered by
  document. A version is kept each time either one is written, whoever wrote
  it: a generation, a hand edit, a restore. Open one to read it, restore it to
  make it current again (which archives what was current, so a restore is never
  a loss), or delete it for good.

Both are generated from your sources and both are shown as a diff against what
is live; nothing changes until you apply it, and each document is applied on its
own.

Adding a source, or switching one back on, asks you to review an update to the
narrative, made by editing its current text. Removing one — a delete, or
switching it Off — marks both documents stale, and the rebuild starts over from
the default structure instead of editing the old text, so nothing the removed
source said is carried forward. Remove every source and both documents reset:
the facts empty, the narrative back to its default. Until you have reviewed a
removal, the guide's second stage stops counting as done.

One **Update** rebuilds both — they always go out of date together, so there is
one button, one progress view and one review with a tab per document. While a
document is being written it cannot be edited, because the result would
overwrite what you typed; editing unlocks as soon as the new text is ready,
whether or not you have read it yet.

Each tab also carries **Rebuild from Artifacts**, for when nothing changed but
the document did: it starts from the default structure and your current
artifacts, the way a brand-new profile is built, rather than editing the text
that is there. Over unchanged sources that mostly re-words the same thing, so
it is one rebuild a day per document; after that the tab says to add, change or
remove an artifact — which lifts the limit at once, because it is also the only
thing that makes a different answer likely. **Manual edit** opens the same
document for you to write in directly, and never spends a generation.

There is deliberately no blank-markdown escape hatch: the intended path is to
import material and correct what came out.

A narrative can exist and still say nothing — headings over "No current projects
listed" — which is worse than none, because the page would greet visitors and
every question the generator writes would be about the blanks. A profile whose
narrative is mostly absence statements is treated as ungrounded: the chat does
not call the model at all, and answers with a fixed note naming the owner and
linking the contact form.

## The chat a visitor has [chat] → https://talenthunt.work/about#chat

The visitor side of a profile, with no account required.

**Starter questions.** A fresh chat opens with a greeting and four question
chips drawn from a pool generated for that person (see [starter
questions](#starter-questions)). After any answer, one pill — *Ask more about
{name}* — opens three more. The pill retires when fewer than three unused
questions are left. The four are seeded from the session id, so they do not
shuffle under the cursor on a re-render, and nothing about the rotation is
stored.

**What the owner sees is the visitor's choice.** A picker in the composer offers
four levels: **Everything** (full transcript, job description and your name),
**Recap + job description**, **Recap only** (the summary, not your words), or
**Private** — the owner sees your role and your city, never your name and never
your words. The setting is enforced where the data is: a level below
*Everything* means the transcript does not leave the server for the owner at
all, and choosing *Private* part-way through a conversation deletes the turns
already written rather than hiding them, after a confirmation that says so.

**A job description can be attached** and the chat answers against it — see [job
descriptions](#contexts). A conversation can also be opened with a JD already
pinned via `?jd=<id>`; the link itself is the capability.

**Contacting the owner** is a form in the chat, and it requires an account: the
message is sent from our domain with your verified address as the reply-to, and
the address is taken from your account rather than from the form.

**Other affordances:** export the conversation as Markdown, share it, save the
owner's contact card as a `.vcf` file, and resume an earlier conversation from a
`?session=` link the owner can hand out.

**Sign-in nudges.** Reading and chatting are free and anonymous. Sign-up is
asked for at specific moments — a few messages in, attaching a JD, exporting,
setting visibility, contacting the owner — and skipping is always possible until
the message count runs well past the first ask.

**Rate limit:** 30 chat requests per IP per five minutes, and four recap
generations per person per hour.

## Starter questions [starter-questions]

The chips are generated per profile, not per visit, and stored as a pool of up
to eight. Generation is two model passes, because a model asked to write
questions and judge whether its own questions are answerable is not a good judge
of the second part: a writer produces twelve candidates, a mechanical filter
drops any that omit the owner's first name or use a pronoun for them, and a
second model checks each remaining one against the narrative.

The rules those prompts enforce are worth knowing, because they are product
decisions rather than style:

- **Answerable means the document states the answer**, not that it mentions the
  subject. Questions about motive ("what made X decide") and about the future
  ("what is next") are named as failures — a CV-shaped document records what
  happened.
- **Every question names the owner**, and no question uses a pronoun for them. A
  profile does not state anyone's gender, and a question that guesses one
  guesses on the owner's behalf.
- **Nothing the owner would have to defend** — layoffs, departures, disputes,
  gaps, failures.
- **Nothing about a job application**, including the fact that one exists.
  Application material is usually attached as a private artifact, which hides it
  from the card but not from the agent.

A profile with no generated pool falls back to eight generic openers with the
owner's name filled in. A profile whose narrative fails the grounding test gets
no chips at all.

**A job description retunes them.** The pool is about the person, but a visitor
who arrived holding a role wants to ask about the fit. So whenever a job
description becomes the chat's context — pasted or uploaded, opened through a
`?jd=` link, or picked from the library — the chips are replaced by three or
four written against that description instead. They still name the owner, and
they are written fresh each time rather than cached: an edited description gets
questions about what it says now, not what it said the first time anyone opened
it.

Nothing waits on this. The page renders with the profile's own chips and swaps
them a moment later, and if the generation fails or comes back empty the profile
chips simply stay — a visitor who never saw the swap has lost nothing.

The swap replaces the pool rather than adding to it, so the *Ask more about
{name}* pill hands out the remaining JD questions too. Three or four of them is
about one reveal, which is the intended shape: once the conversation is about a
specific role, the person-shaped openers are no longer the questions worth
offering. Starting a new conversation brings the profile's own pool back.

The grounding rule is checked first and is not overridden: a profile whose
narrative says nothing shows no chips, with or without a job description.

## Importing from LinkedIn [linkedin-import] → https://talenthunt.work/about#linkedin-import

Signing in with LinkedIn pulls the public profile — positions, education,
headline — and the avatar, with no copy-paste. The avatar is copied into our own
storage, because LinkedIn's CDN URLs expire in about thirty days.

The import **only adds**. Nothing that came from anywhere else is overwritten,
and the imported payload lands as an artifact, so it can be edited, hidden or
replaced like any other. Anyone who signed up with email or Google can trigger
the same import later from Artifacts.

If two accounts claim the same LinkedIn identity, both are flagged for manual
review rather than resolved automatically.

## Job descriptions [contexts] → https://talenthunt.work/about#jd

A job description turns the same chat into two different tools:

- **As the owner** — compare your profile against the JD, see what is missing,
  get told which artifact would answer it.
- **As a reader** — paste the role you are hiring for and ask how this person
  maps to it; the answer cites the material it used.

JDs are saved to a library, so a description is pasted once and reused. Upload
accepts the same document formats as artifacts, alongside plain paste, and a
detector fills in the title and company. The active JD shows in a sticky bar
above the chat, and its full text is one click away. Studio → Job descriptions
manages the whole set, including the ones visitors brought to your chats.

**A page and a listing, if you want them.** When you add a job in Studio → Job
descriptions, the posting is checked first. Then two options unlock:

- **Create a page for this job.** It gets a public page at
  `https://talenthunt.work/fit/{id}` that anyone can open and search engines can
  find. Off by default, and your last choice is remembered.
- **List it on the job board.** It is sent for review, and once approved it
  appears on [the job board](#job-board) and in the JobHunt channel. A listing
  needs a page and at least three requirements in the posting.

A job saved without them can get both later from its card, which also has Share
(link, social apps, QR code) and Open page.

Attaching a JD requires sign-in; opening a chat through a `?jd=` link does not.

An attached JD also retunes the chat's opening chips to that role — see [starter
questions](#starter-questions).

## Fit check against a job [fit] → https://talenthunt.work/about#jd

A fit check takes one job posting's requirements and walks them against a
profile, one line at a time: what the profile covers, what is thin, and what a
recruiter would not find at all. There is no score and no match percentage. The
gaps are named, and each one points to [Artifacts](#artifacts), where the
missing evidence gets added.

**Where it runs.** In any chat with a job description attached, for anyone
signed in:

- **The owner** gets their own check, which ends on what to add.
- **Anyone else** gets a reader's copy, written in the third person, which ends
  on questions worth asking in the chat and a way to write to the owner.

Anonymous visitors do not get a check.

**The vacancy link.** A job description has a page at
`https://talenthunt.work/fit/{id}` only when one was created for it: by its
owner, or by the JobHunt channel for every job it posts. Other saved job
descriptions have no page. The page shows the role, the company, the posting's
full text and a button to check your own fit. Someone signed out signs up
first. Someone already set up goes straight to their own page with the vacancy
attached (`/{username}?jd={id}`), where the check runs. The page's owner sees
the page itself instead.

**Sharing with a vacancy.** The share dialog can send your page with a vacancy
attached, so the person opening it starts from their own posting.

**JobHunt.** The [JobHunt Telegram channel](https://t.me/JobsChannel_1) posts
jobs, and each post carries a fit link for its vacancy.

## The job board [job-board] → https://talenthunt.work/about#job-board

`https://talenthunt.work/job-board` lists jobs that each come with a fit check.
Every card shows the role, the company, where and how you would work, when it
was posted and a short description. "Details" shows the key facts. Opening a
card leads to the job's own page on the board, with what the job asks for and
the full job description, where you can share it or check your profile against
it. The job page keeps your filters, and "Back to roles" returns to the list.

**Where roles come from.** Nothing is scraped from job sites. Roles are added
by the JobHunt curators or posted by members. Every posting is read by AI first,
to check that it is a real job and to pull out what it asks for, and a member's
role is reviewed by a person before it is listed, usually within a day.

**What is listed.** Jobs the JobHunt channel posts, and jobs members sent for
review and an admin approved. A job needs at least three requirements to be
listed. A job leaves the board on the closing date the posting gives, or 90 days
after it was posted.

**Finding a job.** Filter by workplace, location, function, industry, seniority,
job type and date posted, or tell Copylove what you are looking for in plain
words.

**Posting a role.** "Hiring? Post a role free" on the board, or
`https://talenthunt.work/job-board/post`, takes a link to the posting, its text
or a file. No account is needed to start: you see the role as it will look, can
rename it, and choose whether to list it on the board. The account comes when
you send it — sign up, and the role goes to review right away. It gets its own
page at once; the listing follows the review, usually within a day. You can
also add a job in Studio → Job descriptions and choose "List it on the job board".
How the board works, for both sides, is at `https://talenthunt.work/job-board/how-it-works`.

**Limits.** Without an account you can read two postings an hour; after that
you are asked to sign up. A member can create 5 job pages an hour, 10 a day and
20 in 30 days, whether through "Post a role" or Studio. Whenever a limit stops
you, "Ask for a higher limit" opens the contact form with the request type
"Raise a limit".

**JobHunt on Telegram.** The channel posts new roles as they are published, each
with a link to its fit page. It works instead of a subscription to updates:
follow it, or send it to someone who is looking.

**For machines.** Each job's page carries schema.org JobPosting markup and is in
the sitemap while the job is listed. The board is also available as JobPosting data at
`/job-board/jobs.json`, as an RSS feed at `/job-board/rss.xml`, and as
`/job-board/llms.txt`.

## Chat history and recaps [chat-history] → https://talenthunt.work/about#chat-history

Studio → Chat history is every conversation, in three groups: **Chats with me**
(visitors), **Chats by me** (conversations you had with other people's
profiles), and **My self-audit** (your own tests of your own page).

A conversation opens with a card naming who it was with — the same collapsed
profile card the rest of the product uses — then a line of facts about the chat:
when, how long, how many messages, how many of the visitor's questions were
typed rather than picked from the suggestions, and the visibility the visitor
chose. "How long" is time spent, not elapsed time: a break longer than ten
minutes is somebody who left, and it is not counted. The job description, if
there was one, folds under that line. Then the **recap**, then the transcript.

**A visibility setting binds the page owner, not the person who wrote the
chat.** In "Chats by me" and "My self-audit" you always see the whole
conversation, whatever you chose to show the owner of the page you were on.
What a setting withholds, it withholds from them.

Recaps are generated on demand and stored, so a conversation is summarised once.
Generation happens on the server, which is what lets a *Recap only* or *Recap +
job description* conversation still have one: the owner never receives the
words it is made from. Regenerating waits until at least two further questions
have been asked — a recap of a conversation that has not moved would be the
same recap. They are written differently depending on whose side you are on —
owner, visitor, or your own self-audit.

A visitor who chose **Private** appears as a role and a city with no transcript
and no recap, because nothing was kept to summarise.

Each conversation carries a link that continues it, which is what the
notification emails point at.

## Statistics [statistics] → https://talenthunt.work/about#statistics

Studio → Statistics answers one question: what did other people do with your
page. Four numbers over a window you pick — profile views, visitors who
chatted, chats started, questions asked — each with how it changed against the
equal window before it. Click a number and the chart below plots it: one bar
per day up to thirty days, one per week beyond that.

The window is yours: three presets (last 7 days, last 30 days, last 12 weeks),
three calendar ones (this week, this month, this quarter), or any start and end
you pick. **Twelve weeks is the limit in both directions** — the longest range
and the furthest back you can start.

A view is one visitor opening your page, counted once a day however many times
they reload. "Visitors who chatted" is distinct people, so three conversations
by one person count once — which is why it is never the sum of the bars.

**Your own visits and self-audit chats do not count, and neither does traffic
from our team.** These numbers are other people. That is a change from how the
section behaved before: it used to count every session including your own, and
on production that made the chat count about thirty percent too high.

Numbers are per brand: a talenthunt page's statistics never include what your
Work Identity page did.

There is no contact-reveal number, because nothing reveals a contact for tokens
yet. It will come back when that ledger does.

## Sharing [sharing] → https://talenthunt.work/about#sharing

_Status: partly built — the limits are stated in this section._

Studio → Sharing exists because a link travels badly on its own. Five pieces:

- **The link**, with a Copy that adds campaign parameters so landings can be
  attributed.
- **The channels** — WhatsApp, Telegram, LinkedIn, X and email, each opening
  with the intro text already written and the link already tagged.
- **A QR code** carrying your photo and the wordmark in the middle, downloadable
  as a PNG for a slide, a badge or a business card.
- **Intro templates** for LinkedIn and DMs — editable, first-person, saved as
  you type.
- **An email signature** — a real formatted block with name, title, a call to
  action and a tagline, not a raw URL.
- **Resume blocks** — three variants to paste into a CV, for the copy of the
  link that travels as a file.

The same channels and QR are in the Share dialog on the public page, which the
profile card's bottom bar opens at any width, and the chat header opens too
once the chat is on screen. Where the browser has a system share sheet, a
"Share…" button offers it.

**"Embed on my site" is visible but not built** — it is marked coming soon in
the interface, and waits on a card-plus-chat widget.

## Account [account] → https://talenthunt.work/about#account

Studio → Account holds what belongs to the person rather than to one profile,
across four tabs.

**Name & username.** Your sign-in email, your display name, your username, and
an optional auto-derived long username. Also the danger zone: deleting an
account sends the team a request, because deletions are handled by hand for now.

**Contacts.** Email, phone and messenger channels, each with its own ladder per
brand: **Off**, **Reveal** (hidden until the visitor asks) or **Always**. One
channel can be marked primary, and up to four become the card's quick-action
buttons. Today Reveal hands the value over exactly as Always does — the gate is
built but deliberately not enforced while traffic is low.

**Identity verification.** Run through didit, our verification partner, in a new
tab: a government-issued photo ID, about a minute. Only the verified status is
ever public — never the document details. Where the partner is not configured,
the tab says verification is coming soon.

**Billing.** A shared payment instrument for both brands. Nothing is charged
today (see [pricing](#pricing)); the tab reports whether a card is on file.

Social links are a hidden section here for now — they are ordinary link
artifacts tagged as social, and they will resurface with Artifacts' own socials
type.

## Prompts (admin only) [prompts]

Studio → Prompts is a browser and editor for the system prompts — the chat's own
prompt, the narrative and facts generators, the starter-question writer and its
checker, and the per-brand variants of each. The prompt rows in the database are
authoritative; the repository holds no copy of them.

Only accounts with the admin role see the section, and the endpoint behind it
re-checks that on every call — a hidden nav item is not access control. It is
listed here because it exists, not because it is a feature anyone else can use.

## Your URL and changing your username [usernames] → https://talenthunt.work/about#slugs

Your public URL is `talenthunt.work/{username}`. Usernames are 2–30 characters,
lowercase letters, digits and single dashes; six characters and up are
self-serve, and 2–5 are premium.

Two timers protect recruiter links that are already in the wild:

- **90 days between free renames**, counted from your last release.
- **365 days of reservation on a released username**, so nobody else can pick it
  up and inherit traffic meant for you. You can always take your own one back.

The namespace is shared with Work Identity: a username claimed on either brand
is unavailable on the other. Some names are permanently reserved — app routes,
brand names, and a handful of terms we do not sell.

## Premium usernames [premium-handles] → https://talenthunt.work/about#premium-handles

Short usernames are a separate paid path, and a reservation rather than a
rename — so they are not subject to the 90-day cooldown. Early-bird prices are
half the launch price and are locked when reserved: five characters $10, four
$50, three $250, two $2,500. Two- and three-character handles carry a bundled
Pro-Recruiter plan on top.

Approval is manual. Requesting one holds the name for ten days while it is
reviewed.

## Emails talenthunt sends you [emails] → https://talenthunt.work/about#emails

Three, all to owners, none to visitors.

**Welcome**, once, right after signup: your live profile link and four steps —
complete the card, add your career material, test the chat, share it.

**Your first visitor**, once ever, when a real conversation happens on your page
for the first time. The questions are the payload: the mail carries what was
actually asked, up to five questions, with the visitor's name when they shared
it.

**A daily digest** after that, one message per closed day, listing the
conversations that qualified. A conversation qualifies at two visitor turns of
which at least one was typed rather than tapped from a chip — one "hi" is not
news.

The first-visitor mail and the digest also carry a short product tip at the end —
one thing your profile can do, with a link to the screen that does it. At most one
tip a week, never the same one twice, and the welcome mail carries none because
its four steps already say what the first tips would have. A tip is a tail, not
the message: when there is none left to send, or sending one would repeat the
topic you just had, the mail goes without it.

Times are printed in your own timezone, reported once by your browser when you
open the Studio, and in UTC when we do not know it. Nothing runs while a visitor
is talking: a worker sweeps closed windows, so notifications never cost the
reader latency. Our own team's visits are excluded from these emails,
and from your Statistics too.

Every one of them ends with the same footer: the wordmark, a line offering
Copylove for questions about the product, the address the mail went to and why,
and who sends it — WIP Solutions Inc., a Delaware corporation.

**There is no unsubscribe link yet, and the footer says so** rather than showing
one that does nothing: "Don't want emails like this — reply 'Stop'." A reply
reaching us is read by a person and acted on by hand. That is a real gap, not a
design choice, and it is the thing a weekly mailing would have to close first.

## Privacy and data [privacy] → https://talenthunt.work/about#privacy

You decide what is public. Anything at Link or Download on the card is visible
to anyone with the URL and indexable by search engines and AI crawlers — that is
the point of the product. Anything Off or Private never leaves the server as
content: the public read filters it in the database.

A profile is public only once it is listed and consent is recorded; the sitemap
and the public card follow the same rule, so an unfinished signup advertises
nothing.

Conversations are stored so owners can review what was asked, subject to the
visibility the visitor chose. Data is not resold and is not used to train
third-party models. Product analytics run on PostHog at the scope you would
expect — page views and basic interaction events. Verification documents are
never shown to anyone; only the resulting status is.

Local development and preview deployments write no conversation rows and send no
mail, so testing the product cannot file a conversation somebody never had.

## For AI agents [agents] → https://talenthunt.work/about#agents

- `GET /{username}` — server-rendered HTML with Schema.org `Person` JSON-LD in
  the head: `name`, `url`, `mainEntityOfPage`, plus `jobTitle`, `description`,
  `image` and `sameAs` (LinkedIn) when available.
- `POST /api/chat` — body `{ "slug": "<username>", "messages": [{ "role":
  "user", "content": "…" }] }`. The reply streams back as plain text. No auth is
  needed for a public profile. Rate-limited to 30 requests per IP per five
  minutes. `messages` is the whole conversation so far; a single-message array
  is a valid first turn.
- Include a job description in the message content to get JD-aware answers.
- `/robots.txt` names the major AI crawlers and search agents explicitly —
  OpenAI, Anthropic, Perplexity, Google-Extended, Applebot-Extended, CCBot,
  cohere-ai, Bytespider, Diffbot, Amazonbot among them — and disallows only the
  signed-in and API paths. It is the authoritative list.
- `/sitemap.xml` lists every public profile plus the public pages.
- Answers are grounded in the owner's own material. A profile whose owner has
  written nothing answers with a fixed note and a contact link rather than a
  guess.

## Pricing [pricing] → https://talenthunt.work/about#pricing

talenthunt is always free for job seekers, with no card required, and that part
does not change. Usage-based pricing for recruiters comes later; while the open
beta runs it is free for them too. Accounts are not split by role, so the free
tier is built around job seekers first — and signing up now locks in a
substantial discount on whatever is introduced later. The only thing priced
today is a premium short username, and that is a one-off reservation reviewed by
hand.

## Tech stack [tech]

Next.js (App Router) · React · TypeScript · Tailwind CSS · Supabase (auth,
Postgres, storage) · Anthropic Claude for every model call · Resend for mail ·
PostHog for analytics · didit for identity verification.

## About talenthunt and the name Copylove [talenthunt] → https://talenthunt.work/about#talenthunt

talenthunt is a visibility platform for job-seekers, where verifiable work
history, references and reputation live. Copylove is the agent that acts as the
conversational front door to one person's story, for human recruiters and AI
sourcing agents alike.

The name is a memorial. `/copylove` tells that story and is not a product page.

Work Identity is the sibling brand on the same engine: one account, one username
namespace, one shared pool of material, two profiles. Only talenthunt is live
today.

## Getting help [help] → https://talenthunt.work/about#contact

The support form is in the site footer and on the account page, and it reaches a
human. Replying to any email we send works too.

## Key URLs

- Home: https://talenthunt.work/
- How it works: https://talenthunt.work/about
- FAQ: https://talenthunt.work/#faq
- For recruiters: https://talenthunt.work/for-recruiters
- Why the agent is called Copylove: https://talenthunt.work/copylove
- Legal: https://talenthunt.work/legal
- Sitemap: https://talenthunt.work/sitemap.xml
- Full brief: https://talenthunt.work/llms-full.txt

## Featured profile

- talenthunt founder: https://talenthunt.work/md