What you are asked to agree to

What changed since you last agreed to this

If you accepted version 7, one thing has been added to what your Search Team does as you.

It tells you the moment an employer asks to interview you. When a reply it reads is an interview request, it emails you one line — who, for which role, and a link to the page — from your own address to your own address. If you have added another address on your profile page and proved you hold it by entering a code sent there, a copy goes to that address too. That is the only thing it sends there, and only you can add one.

If you accepted version 6 or earlier: version 7 withdrew a permission rather than adding a verb. We no longer ask Google for access to any mailbox; signing in with Google tells us only who you are, and your job-search mailbox is the one we make for you, the only mailbox your Search Team reads from or sends from. Version 6 added that your Search Team can make an account for you on an employer's hiring system, for an application you have approved, under your own job-search address, with a password it makes and keeps sealed. It is described in full under "About applying" below.

First, what your Search Team actually is

Your Search Team is software. It is a set of AI agents — large language models — that read your mail, draft your messages, and act on your behalf.

No person is reading your mailbox.

We call it a Search Team everywhere else because that is what it does for you. Here, where you are deciding whether to hand it authority, you should know exactly what it is.

What you are authorizing it to do

This is the whole list. There is no "and similar activities" at the end of it.

• Read your job-search mailbox. To find employer replies, job alerts, and the sign-in codes that employer portals send.
• Send email from your address, as you. Follow-ups and replies to employers. These arrive looking like you wrote them, because as far as the recipient is concerned, you did.
• Send your search to Anthropic, so a model can read it. Your profile answers, the job postings being considered for you, and the messages in your job-search mailbox go to Anthropic — the company that makes Claude, the model your Search Team thinks with.
• Send an application in your name, to an employer you approved it for. By email, with the exact resume and the exact letter you read and said yes to, from your address and over your name.
• Fill in and submit an employer's own application form, for a job you approved it for, making an account on that employer's system when it will not take the application without one. The same exact resume and letter, and answers taken only from what you have told us. Where the form asks something your profile does not answer, it stops and asks you rather than guessing. The account is under your own job-search address, with a password made and kept sealed so that no model and no person sees it.
• Send what you write to your Coach to Anthropic, with what it needs to answer. The Coach is the part of your Search Team you talk to. It can tell you where things stand and ask the rest of the team to do something; it cannot send, submit or approve anything.
• Email you a morning report, from your own address to your own address. One message a day: what needs you, what went out, what came back, and what your Search Team did on its own. Nothing in it goes to anyone else.
• Email you the moment an employer asks to interview you, from your own address to your own address, and send a copy to any other address you have added and verified. One short line naming the employer and the role, with a link to the prospect's page, sent within the hour of their reply being read. A copy goes only to an address you typed in on your profile page and proved you hold by entering the code sent to it; nothing else is ever sent there, and you can remove it at any time.
• Approve applications for you, only if you have asked it to, up to a number a day that you set. A standing yes is off until you turn it on, and you can turn it off at any time. When it is on, your Search Team says yes on your behalf to applications whose words are already written and waiting for you: best fit first, at most your number in a day, and never one that would wait behind another application at the same employer. Each one is marked on its page as approved for you, and you can withdraw it before it goes.

That is what Fried Mango can do today. It is deliberately short.

About applying

You approve every single application, one at a time, before it goes. Not a setting you turn on once, and not "applications like this one". You read the exact letter and open the exact resume, and you say yes to that one. If either is rewritten afterwards your yes no longer covers it, and we ask again.

We do not decide a job is worth applying to and then apply. Your Search Team reads postings and decides which are worth your time; it writes what it would send; then it stops. The sending is your act, performed by us on your instruction.

The resume an employer opens is the file you opened. It is made once, when the words are written, and stored exactly as it is. We never rebuild it on the way out — so what lands in their inbox is the document you looked at, not a fresh version of it.

By email where there is somebody to write to, and through the employer's own hiring system where that is the only door. Either way you read the exact words and said yes first. On a form, every answer comes from what you have told us; a question your profile does not answer stops the application and comes to you, and nothing is submitted until you have answered it. Some employers email a code before their form will accept an application. Your Search Team reads that code from your job-search mailbox, which is the mailbox permission you have already given, and nothing more.

Where an employer's system requires an account before it will take an application, your Search Team makes one for you. It uses your own job-search address, the one made for you, and never any other; it makes a password itself and keeps it sealed, so that the model working the form never sees it and no person here can read it; and it makes at most one account per employer system. If the employer emails a code to confirm the address, your Search Team reads that code from your job-search mailbox, as it already does for a code sent after submitting. It signs in to that account only to make or finish an application you have approved, and uses it for nothing else. You are told, on the prospect's page, when an account was made and where. If an account under your address already exists there and we hold no password for it, your Search Team stops and tells you; it does not guess passwords, reset them, or make a second account under a different address.

Your Coach is the part you talk to. It reads what is on file about you and where your search stands, and answers you. It can ask the rest of your Search Team to look for jobs, write an application for you to approve, get you ready for an interview, or read your inbox. It cannot send anything, submit anything, or approve anything, and it cannot change your mind for you: approving is still you, on the prospect's page, reading the exact words. When you tell it something about yourself it is kept on your profile, marked as something we worked out from what you said, and nothing uses it until you confirm it there.

One thing we will not do without asking you again: apply to more than one job at an employer where you are already talking to someone. Applying past your own recruiter is worse than doing nothing, so those come to you as a conversation to have rather than an application to send.

About the model

Anthropic is a supplier, not a partner. They run the model; they do not get a say in your search and they are not told who you are beyond what is in the text we send. We chose to name them rather than say "an AI provider", because a name is something you can look up and a category is not.

What they may do with it, in their words and ours:

• They do not train models on it. Their commercial terms for the interface we use say so directly, and that is the basis on which we send anything at all.
• They delete it within thirty days. That is their standard retention for this interface. It is not zero — we have no arrangement with them for zero — so for up to a month a copy of what was sent exists on their systems as well as ours.
• One exception, which we would rather you heard from us. If something we send trips their abuse detection, they may keep it for as long as two years while they look into it. We do not expect that to happen to a job search. We are telling you because a promise of thirty days with a footnote is worse than a plain description of both.

If we ever add a second model company, that is a new version of this document and we ask you again. Not a notice, not a footnote — the same conversation we are having now. That is the reason this names one company instead of leaving room.

What it will not do

• It will not apply for a job you have not approved, and it will not treat one approval as permission for the next. A standing yes is you approving in advance, up to a number a day; it is not permission for anything past that number or past the words on file.
• Nothing goes out until you have read the exact words and said yes, unless you have set a standing yes. Not a summary. Not "an email like this one". The actual text, the one that will be sent. With a standing yes the words that go out are still the exact words written for you to read; what changes is that your Search Team says yes to them for you, marks each one, and you can withdraw it.
• It will not touch any mailbox but the one made for you. Not the account you sign in with, and not any other.
• It will not put what it reads into a model that serves anybody else. What you tell it and what it finds makes your search better, which is the whole point of it. None of it trains a general model, none of it improves another candidate's Search Team, and Anthropic does not train on it at all.
• It will not send your credentials anywhere. The model never receives the key to your mailbox, your passwords, the passwords it makes for accounts on employers' systems, or anything that would let it or anyone holding it sign in as you.
• No person here reads your mailbox — not for support, not for debugging, not out of curiosity.

Why your job-search mailbox is one we made for you

Because this authorization is real, and it should be as small as it can be.

A mailbox made for your job search contains your job search. Your everyday account contains fifteen years of your life. We cannot read that, and we do not ask to. Signing in with Google tells us who you are and nothing more; the mailbox your Search Team works out of is one made for you, on our own mail service, and it holds only what the search puts there. A Search Team reading job postings all day is one bad posting away from reading things that are nobody's business, and the way to keep that from mattering is for the mailbox it can reach to contain nothing else.

That reasoning got stronger with version 2, not weaker. What we can read is also what we can send to a model, and the smallest mailbox is the smallest version of both.

It protects you the other way too. If something ever goes wrong with the job-search mailbox — our fault or nobody's — you have lost a mailbox made for this, not the address your bank, your doctor and your family use.

This is a version 0

Fried Mango is new. It works, and it is not finished. What it can do will grow.

Two different things will happen as it grows, and we promise different things about each:

The second one is the promise that matters. If we ever change what it does as you, we will ask — not tell. A new verb means a new version of this document and a fresh acceptance, and until you accept it, the new thing does not happen.

That is why the list above is short and specific rather than broad and comfortable. A broad authorization would let us grow into it quietly. This one cannot. Versions 3, 4, 5 and 8 exist because of exactly that: sending an application, then applying through an employer's own system and talking to a Coach, then a morning email and a standing yes, then an email the moment an interview arrives, were new verbs, so you are reading this instead of being told afterwards.

What is coming, so you are not surprised by the shape of it: signing in through the same service that holds your job-search mailbox, so that Google is not needed at all. That changes how you get in and nothing about what is done as you, so you will be told rather than asked.

If you change your mind

Tell us, and we will stop and delete what you would like deleted. Everything your Search Team does runs through the mailbox made for you; disconnecting it stops the reading and the sending, and we can do that the same day you ask.

Who this applies to

The first seven accounts, and no more than seven. Each one is invited personally.

If you are reading this, you are one of them.

Why there are only seven

Fried Mango is a version 0. It works, and it is not finished.

You are not getting a deal. You are running your actual job search on new software — which is about the least forgiving thing anyone can test anything with, because a missed reply or a badly worded follow-up is not a bug report, it is a job you did not get.

Seven is small because seven is how many people can be paid real attention. If something goes wrong you will be talking to a person who knows your account, not to a queue.

What we get from this is the only thing that actually proves a job-search tool works: whether it holds up across real searches, run by people who are not us. That is worth a great deal, and the free period is what it costs. It is not a promotion.

What will change is the product — more of it, and better. What it does on your behalf is a different matter, and is covered separately in the authorization you accepted alongside this: if that list ever grows, we ask you again rather than telling you.

What you pay right now

Nothing.

How long that lasts

At least two months, counted from the day you accept these terms — not from the day you signed in, and not from the day you were invited.

"At least" is the honest word. Two months is a floor, not a deadline. It may run longer, and if it does you will not have to ask or chase anyone: it simply continues until we tell you otherwise, and we cannot tell you otherwise without giving you notice.

What happens when it ends

The standard rate is $100 per month, billed up front.

Before any of that happens, we will ask you for payment details. We do not have them, we are not holding them, and nothing can be charged to you until you give them to us and say yes.

What you will get before you are ever charged

Thirty days' notice, in two places:

1. An email, to the address you signed in with.
2. A banner on your dashboard, which stays until you have dealt with it.

Two places rather than one because an email can go to spam and a banner can go unread, and neither of those is your fault. Thirty days is enough time to decide, to export what is yours, or to leave.

Because the free period is a minimum of two months and the notice is thirty days, the earliest you could hear from us about money is about a month in. If you have heard nothing, nothing is happening.

One thing this does NOT promise

Being early does not lock in a price. If the standard rate changes later, you pay the standard rate.

This is written down because it would be easy to leave vague and let you assume otherwise. What being early gets you is a long free period, honest notice, and somebody's actual attention. It does not get you a permanent rate, and pretending otherwise now would only be a worse conversation in a year.

If you leave

You can stop at any time, and there is nothing to cancel while you are not paying. Your data is yours; ask and we will give you a copy, or delete it.

If this document changes

A material change — anything about price, duration, or notice — requires your agreement again. You will be asked; it will not happen quietly, and the old version stays on record as the one you originally accepted.

Fixing a typo is not a material change.