Your first launch and business brief

Your first launch

Guided onboarding shows five review steps: Starting point, Business brief, Customer profile, Research, and Draft review. Completing them opens your workspace. Save and exit keeps an unsubmitted website or concept on your account, so you can close the tab, return from Today on any device, and resume at the same step. The page also saves it a moment after you stop typing. This early input is not a confirmed business brief and never overwrites one. If the account copy cannot be saved, the page stays open and says the input is held in that browser tab only. Start with your website or a short description. Review the brief and use AI Assist to fill or refine any section from everything you have entered; two fields on the brief, what is available today and what the team should avoid promising, shape the market and draft prompts so a concept-stage business gets research aimed at learning rather than a sale. After confirmation, the assistant proposes customer profiles and an addressable market estimate with visible sources and assumptions.

In draft review, Save and exit saves each changed outreach draft before returning to Today. If a save fails, the page keeps the remaining edits for retry. Saving a draft does not send it or place it in a mailbox.

Choose a profile to research up to ten companies and five contacts. An optional list of accounts or markets to exclude keeps specific domains or names out of research; an excluded company still shows on the research screen, marked as such. Exclusion is a decision, not a research flag: the same rule holds the account at contact selection, before a draft is written, and again immediately before a draft is placed, and it applies to the recurring plan as well as this launch. Your own records add rules too: a contact who opted out or bounced, an account that is already a customer, and an account with an open opportunity are held with a sentence naming the record. The research step lists the recommended accounts as one table with four columns. Account shows the name and domain. Why it fits gives the reason, the research and the sources that establish the company. Eligibility shows an Eligible, Excluded or Not confirmed badge, when the account was researched, and what research it used, in AI calls and vendor credits, with a link to the receipt for each paid lookup. Action holds Exclude account or Restore eligibility. Below the table, the Contact evidence panel lists the contacts found at each account, grouped under it, beside the Research request panel. Each contact shows the person, the employer and the mailbox as three separate lines, because a vendor naming a real person says nothing about whether mail to that address arrives: the mailbox reads Deliverable only after a current check, and Unknown, with the reason, for an address nobody checked, a domain that accepts any address, or a check that has expired. A row researched before research use was recorded says Not recorded rather than showing a zero, and the calls that built your market serve every account, so they are counted in the launch total under the table and not on a row. The Research request panel names any paid request whose outcome is still open, with its receipt. Before any contact is found, the Contact evidence panel says so. Restoring lifts only a rule you set yourself, and a row held by your records says so. On a phone each account becomes a stacked card with the same fields. A person a rule holds never costs an enrichment credit, and a draft held before placement says why on its card. Saved contacts are checked first. You can approve up to five enrichment credits from your connected Apollo account. Existing Clay or ZoomInfo contacts imported into your workspace can be reused; direct research availability differs by provider. Your active seat and account budgets apply to AI work. The Activate workspace step shows readiness per role, what the first workflow produced and what comes next, plus the seat.

What the seat card can say, and why each one is different. Active, with nothing left to do. Setting up your workspace, once the payment has settled and your confirmed brief is being turned into a first launch. Payment pending, while the bank is still deciding. Checkout did not complete, which is the one Stripe cannot tell you itself: a declined card is sent back to the same address as a settled one, so the page used to say pending forever and offer a Check again over a charge that had already been refused. That card now says what happened and offers another card instead. Payment received, for the case where your money moved and the seat has not reached your account: it says so and tells you not to buy a second one, where it used to show a buy button. Your workspace could not be set up, which used to render as "being set up" while Try again reproduced the failure. Cancelled, with nothing charged. And, when a seat cannot be sold at all, a sentence saying so rather than a button that fails when you press it. The state is said once, in a boxed notice at the top of the page: blue while a seat is still needed, amber while payment or activation is settling, red when payment or setup failed, and green when your workflow is ready to start. The Workflow status card underneath keeps the facts (payment, seat, workspace) and the actions, and after a check the answer appears in that notice, not repeated on the card.

The page opens on the state the payment is in. Without a seat, the page reads the payment notification we last received for your checkout before it draws anything, so a declined card opens on Payment did not complete with Try another card, a payment still settling opens on Payment is still processing with Check payment status, and a payment that went through with no seat yet opens on Payment received; activation is retrying with Check activation. You no longer have to press a button to find out. Each of these keeps your starting point on screen, the website and description you entered, beside the card. While there is a seat to buy, the card is headed Activate your seat and lists the price from the live catalog and the first work the seat starts.

We retry activation for you. When a payment has gone through and the seat has not reached your account, or your seat is active and the workspace could not be made, a scheduled job tries again on its own, every ten minutes at first and then less often, up to six attempts over about five hours. The first attempt runs within 30 minutes of payment. The card's Automatic retry line says when the next attempt runs, in your own time zone, and how many have run. You can leave the page; Check activation and Try again still run the same step straight away. The retry uses the same grant and setup as the payment notification, so it cannot give you a second seat or a second workspace. It does not retry a seat you cancelled, and it does not rebuild a workspace from a brief with no company name; add the name and press Try again. If all six attempts fail, the card says the retries have stopped, our team is alerted, and the card links to Support. Do not buy a second seat in that case: the payment is on record and we will find it.

Your workspace is made once, from the brief you confirmed, as soon as your seat lands. Both the payment notification and your own Check again make it, and whichever arrives first wins: a repeated, delayed or out-of-order notification cannot make a second workspace or overwrite the first. It is made for an account that holds a seat, however that seat arrived; no workspace is made without a confirmed brief, or from a brief with no company name, because a launch built from nothing cannot run and would not be able to say why.

A launch is created once per account and nothing deletes it, so if you edit and re-confirm your brief afterwards, creating again tells you that you already have one rather than quietly leaving you on the older version.

The price on the card is checked against Stripe when checkout opens. A mismatch refuses the purchase rather than charging you a different number. Checkout also waits for you to confirm your email address: until you open the link we sent at signup, Add your seat says so, sends the link again, and charges nothing. Confirming the brief without a seat keeps the brief and names the blocker; once the seat lands, research starts from the brief you confirmed.

Progress is saved. You can leave during research and return later, or pause after the current step; the scheduled worker keeps stepping a launch for up to a minute at a time while you are away. A research step can take a couple of minutes. If a call is cut off, it is retried twice before the launch pauses; other failures pause it at once, and the page explains what needs attention and keeps completed work. The launch has a ceiling of 30 AI calls and four web searches per call.

A paused launch with an unresolved research request. If research paused with a paid research request whose outcome is still unknown, the research screen shows a Research request box naming the request key with outcome unresolved, the credit reserved for it, and a Review request receipt link to that request's receipt on Receipts. Check it before starting another request: a second purchase for the same contact is held until the first is settled. With no such request the box says so, and if the check itself could not be read the box says that instead.

Launch authorized. Your confirmed brief is what authorizes the first launch. Once your seat is active, the brief step shows a Launch authorized box with who confirmed the brief and when (read from the confirmed brief version; a value that was not recorded reads Not recorded), the scope (up to ten companies, five contacts, 30 AI calls and five vendor credits, with every email kept as a draft) and the launch key, the one launch this account has, which is created when you start market research. View launch progress opens the running launch. Before a launch exists it is disabled and says why.

A company counts as confirmed once a search result from its own website backs it. Candidates the market step names without that evidence are kept aside, researched after the confirmed ones, and never used for contact research unless their own site is found.

What is known about each recipient, and how. Every contact card keeps three things apart: who the person is, where they work and as of when, and whether mail to the address arrives. A vendor returning an address is not a deliverability check, and a domain that accepts mail to any address proves nothing about the person, so both are marked for review rather than shown as verified. An employer confirmed more than four months ago goes stale, because people change jobs. A contact whose evidence does not stand on its own is not placed. Verify contact checks the address through your connected Apollo account for one launch credit; if it still cannot be confirmed, choose another contact at that account.

Paid research is written down before it is paid for. Each enrichment records the person it is about, the credit set aside and the result. An answer you already bought is reused rather than bought again, and a call that was cut off stays unresolved instead of quietly repeating. An unresolved request blocks a second purchase for the same person until you settle it: a vendor that charged and lost the response has both your money and your answer, and buying again returns neither.

Review each email and its recipient, check the connected Gmail address, then approve placement in that mailbox's draft folder. You send from Gmail yourself. A returned Gmail ID confirms placement. If the outcome is uncertain, check Gmail; the assistant will not risk creating a duplicate.

Coming back to a launch you already have. One launch is created per account and nothing replaces it, so returning to this screen shows what you already have rather than offering a fresh start: whether it is running, paused or finished, and what it produced. Onboarding never resets a workspace to fix itself.

A launch runs the version of the brief it was started from, so work you already approved cannot change underneath you. If you have edited and re-confirmed the brief since, the screen says so and names which sections moved, rather than leaving you to notice it in a draft that uses the wrong company name. The newer version is what your scheduled work reads. Pressing to create a launch when you already have one now tells you that, instead of answering as though a new one had been made.

What is already running. When you come back with work under way, an Already running box sits above every control that starts more. It counts the contacts in your daily prospecting plan (with how many were stopped and how many answered), the sends that started and are still waiting on a receipt, and your active Instantly enrolments. Each count links to where you can see the list: the daily prospecting plan on this page, Receipts and Campaigns. Anything you start from here adds to that work. It does not replace, pause or stop it. If a count cannot be read, the box says so instead of showing 0, and Check again reads it once more.

The detailed business editor retains the source picker and all eight sections described below. A launch uses its own confirmed snapshot, so later edits do not change work you already approved.

Overview

Guided Activation uses the shared workspace typography and panels through setup and saved progress. Progress navigation, Save and exit, and the detailed editor remain available. Payment and provisioning checks retain their existing behavior.

Instead of filling a long form, point us at your context at Tell your roles about the business. The ingest agent reads what you give it and researches outward, then shows you a GTM Context Dossier: company identity, products, founder, voice samples, ICP, and competitor intel. You edit it, then confirm. The confirmed dossier is what the Writer clones your voice from, what Scout uses to find relevant angles, and (since 1 September 2026) what every scheduled role reads before it does any work.

Until then the dossier fed the content products only. Now the same confirmed record is put in front of each of the nine roles on every run, so a pre-call brief, a renewal play or a partner note is written for your company and your buyer rather than for a generic one. Nothing changes about how you fill it in.

It's the "feed the archive" step from the playbook this engine is modelled on. The Writer sounds like you because it's fed your real material.

Start from your website

Paste your website on Your first launch and choose Prepare my brief. The scanner reads the page you give it and up to five more on the same site (pricing, about, products, features, solutions, customers, case studies, docs), each page through the same URL, size and time checks, and saves a draft for review. A description works too. Reading the site, like every AI step after it, needs an active seat; without one the step keeps what you entered, says so, and offers the seat there. If you are signed out when you come back from checkout, sign-in returns you to the same state. If the tab you come back in no longer holds what you entered, the page still says how checkout ended and shows the seat card with Check payment status; it tells you to enter your website again rather than claiming it was kept.

Scanning again after you confirmed. A scan against a confirmed brief writes nothing to it. Each section the scan read differently becomes a card under What your roles noticed on the detailed editor, one per section. Accept adds what the scan found (for company and founder details, it takes the changed fields) and keeps the brief confirmed; decline keeps what you wrote. Both screens tell you the brief is kept and how many proposed changes are waiting.

Before you have an account. The homepage asks for your website and runs the audit on it: it reads up to six pages, drafts your company, buyers and competitors, and writes a sample post, showing each step as it runs. No account is needed. These runs share a fixed daily budget; when it is used up, the page says so, saves your request, and emails the result after the budget resets at 00:00 UTC.

If you used the audit with the same email, its business context can already be your starting brief. The detailed editor shows the audit source and offers Read my site for a fuller scan. It also includes public and private repository sources, Google Drive and pasted documents.

Confirm the brief to prepare your first market. Existing business details stay available to the other roles in your workspace.

Keep prospecting after your first launch

On the final onboarding step, connect Attio, verify your Gmail mailbox and choose Start daily prospecting. The assistant uses your approved ICP to research up to five new companies (three under a lighthouse motion) and five contacts each day, add records to Attio and prepare Gmail drafts. Optional Apollo enrichment uses your connected account. Existing AI and enrichment budgets apply.

The plan keeps the brief it was started with, so a change you confirm later does not rewrite work already under way. When your confirmed brief has moved on, the plan says which sections changed and offers Use the confirmed brief; the next research and drafts read it from then on.

The plan shows its follow-up sequence as one table, This reviewed sequence: each step, when it happens, and what stops it. Starting the plan approves exactly that version. If we change the timing or the stop rules later, follow-ups wait (daily research carries on), the plan says the sequence changed, and Approve this sequence resumes them once you have read the new table. The same table, with its approval state and an Approve sequence control, is on the Dispatch Queue under the draft card's This reviewed sequence scope.

Each contact enters a monthly review cadence. After Gmail confirms you sent a draft, the next follow-up becomes due one calendar month later. Pending drafts stay in place until you review them. Replies, delivery notices and opt-outs stop further drafting. No emails are sent automatically.

The same screen shows Attio sync, Gmail receipts and due dates, and Home shows the same plan every day. Pause the plan or stop a contact on either. Review an uncertain CRM result in Attio before using Retry Attio lookup. The plan pauses new drafting at 25 pending drafts and new enrollment at 500 contacts; provider errors, queue delays and budget limits can delay daily work.

The eight sections

Company identity, products, founder, voice samples, ICP, go-to-market plan, target accounts, and competitor intel. The detailed editor adds a ninth, the narrative, which saves with the company section. The last two are decisions rather than facts about the world, so the scan leaves them to you. Section assistance proposes a value and asks back the one or two things only you know; see below.

The narrative: one belief, argued every week

The detailed editor has a Narrative section under the ideal customer. It holds four things your content keeps coming back to:

• Conviction: one sentence someone could argue with.
• Enemy: what your reader is against. A way of working, not a named competitor.
• Belonging: who your reader becomes by agreeing with you.
• Story bank: up to seven moments from your own work, each with the lesson it proves.

AI Assist on the section drafts a first version from what is already on file: your voice samples, founder background, positioning and product copy, including what the free audit read. It is a proposal. Choose Apply to put it in your draft, then confirm the brief as usual; nothing reaches your roles before that. A story is kept only when it quotes the words in your brief it came from, and the card shows that quote. When your brief tells no story, the slots stay empty with a prompt to add your own. No agent writes a story for you.

Once confirmed, every role reads the conviction, enemy, belonging and up to three stories. The content strategist argues the same conviction all week and changes the proof: your own story, a customer story, data, an opinion or a case study, each brief tagged with its type. The messaging experiment tracker tests six to eight wordings of the conviction and reports which ones people quote back. Morning drafts use the conviction for framing and leave the stories out.

Stories also come from this week's questions in Studio. When you answer one with a moment worth keeping, choose Keep as a story, write the lesson it proves, and send it to your brief. It lands in your draft as a change waiting for review, marked as your answer, and reaches your roles only when you confirm the brief.

The narrative is stored with the company section of your brief, so it is versioned, compared and merged with the rest of it.

Your market motion

The brief asks two short questions, on the first onboarding screen and in the go-to-market section of the detailed editor:

• How exposed is the buyer who signs? Exposed means the product is regulated, replaces a system of record, or produces output the outside world sees. Low risk means it is easy to try and easy to undo.
• Does proof travel in your market? Yes when buyers watch each other, status matters, or a few big names lead. No when each buyer decides on their own evidence.

With the number of companies on /tam, when it has any, the answers point to a motion, shown under the questions with one line saying why:

• Lighthouse: an exposed buyer in a market where proof travels, or a small market (under 200 companies on /tam) where it does. Win a few named accounts the rest of the market watches.
• Signal-led: proof does not travel and the buyer is exposed or the market is small. Approach an account when it shows a reason to move.
• Landgrab: a low-risk buyer in a wide market, or one where /tam has not been counted yet. Reach many accounts quickly.

Disagree, and pick your own under Your motion; your pick is saved with the brief and wins over the answers. Leave both questions blank and nothing changes.

Once the brief is confirmed, the motion sets these defaults:

• Every role reads the motion with what it means for list size, how a first email opens, pace and the next step. Lighthouse adds a time-boxed pilot with named milestones as the step to propose; landgrab adds strict qualification, dropping any account that misses an ICP criterion and saying why. The launch's market and draft prompts read the same lines.
• Daily prospecting researches three new companies a day under lighthouse and five under the other two, as it does today. It uses the brief the plan was started with, so choose Use the confirmed brief to bring a new motion into a running plan.
• Morning drafts for a tier 3 account open from that account's fresh signal under lighthouse or signal-led, the way tier 2 drafts already do, when the person gave no hook of their own. Under landgrab, a draft for an account in a pain segment on /tam opens on that segment's one tested line, and an account whose segment is resting waits for its next round. An account in no segment keeps today's draft.

Prepared fields and AI Assist

Empty writing fields can arrive with suggested text from the current brief. Preparation does not save, confirm or approve anything. Review the text before saving. Your saved profile can supply your signature name; missing identities and unsupported facts remain for you to confirm. Credentials and structured data use setup guidance rather than generated values.

Each writing field has AI Assist for a reviewed edit. Use suggestion applies it to that field, and Undo restores the previous text until you edit it again. Suggestions use your AI allowance. Automatic preparation is bounded to four batches of up to twelve visible writing fields per visit, with a fifteen-minute account and revision cache. Later fields retain on-demand assistance when that limit is reached.

Section assistance

In guided onboarding, open Customer profile and the rest of your context and choose Review this section with AI under one of the eight sections. It proposes a revision from the draft so far, plus web research for the four outward sections (ICP, go-to-market plan, target accounts, competitors). The proposal appears near the top of the brief with its sources. Choose Use this suggestion to apply it, or Keep mine to leave your text in place. The detailed editor keeps its section-level AI Assist, Apply and Dismiss controls.

Where the answer is yours alone, the card asks at most two questions, such as your target this quarter, typical deal size or priority segment. Type an answer and choose Answer and refine in guided onboarding, or Answer and re-run in the detailed editor. The proposal is rebuilt with your answers.

For target accounts the proposal is a list of eight to twelve named accounts that fit your ICP, each with a domain and a timely reason to contact it. Applying the proposal adds them to your list as founder-stated rows; check the research before a role writes to them.

Assist runs on the platform key under the same seat and spend limits as the scan. Company, products, founder and voice are read from your own material and never searched, so a competitor's tagline cannot end up in your brief.

The go-to-market plan is how you sell, the motion, the segments, the stage you are at, the number you are working to, your typical cycle and deal size. A role reads it before it proposes anything, so it does not offer you an inbound play when you run outbound, or a scaled program when you have four customers.

Target accounts are the accounts you decided to go after, one per line as name, domain, why. Only the name is required. The reason is what lets a role tell an on-strategy account from an inbound accident, so give one where you have one.

The five sources

• Public repo (URL), reads the repo's README. No auth.
• Private repo (GitHub), same, for private repos: paste a fine-grained GitHub token (repo read) and we read its README.
• Google Drive folder, point at a Drive folder (paste its link or id) and we read the Google Docs and text/markdown files in it. Connect Drive on the Connect page (read-only), or paste a Drive access token; no connection? paste the text instead.
• Local upload / paste, upload a folder of text/markdown docs (we extract and stage each file's text) or paste your context directly. Works for anyone, no auth.
• Live website crawl, reads your public site. No auth.

How to use it

1. Open /onboarding/dossier and pick a source.
2. Paste the text or enter the URL, then Scan & research.
3. Review every field. Competitor and ICP items are researched, verify them.
4. Confirm this version. The content engine can now write in your voice, and every scheduled role reads this version.

Nothing reaches your roles until you confirm; before the first confirmation the brief is a draft for review.

Your business, safely versioned

The detailed editor places Active business context beside Proposed changes. The active panel reads only the confirmed snapshot; saved drafts and AI suggestions do not replace it. A concurrent confirmation refreshes that panel while keeping your edits available for comparison. If the saved context cannot load, the page says so and offers a reload.

Once a brief is confirmed, the version your team is using and the edits you are making are two different things. The strip above the editor names the active version (a short code, when it was confirmed, and by whom, read from the version record: "by you", or "from a proposal you accepted"). The Active business context panel lists the business, customer, positioning, evidence (the sources the brief was read from) and that version. Each field in the proposed changes says where its value came from: From version and the code while it matches the active version, and Edited by you, not confirmed yet once you change it. A brief that has never been confirmed has no version to point at, so its fields carry no such line. Save draft keeps your edits as a draft beside it; the context your roles read does not move. Confirm this version makes the draft the active version, in one step: it either fully succeeds or leaves the previous version exactly as it was. New research and drafts record the version they were built from, so you can tell which business a piece of work assumed.

Three things can interrupt a confirmation, and the editor names each one. If a newer version was confirmed while you were editing (another tab, another device, a cofounder), the strip says A newer brief is active, the form's heading changes to Compare your changes, and the strip compares the two field by field: each field that differs, the value in the active version, the value in your draft, and who changed it since the version your draft started from. Review merged draft builds a draft on top of the active version. It takes your change where the active version left that field alone, keeps the active value where only the other version changed it, and asks you to choose, per field, where both changed it; until you choose, it saves nothing and says why under the button. The merged draft is saved as a draft. Nothing is activated until you confirm it. If the version your draft started from cannot be read, every difference asks for a choice. You can also load the active version or keep editing your draft; nothing is overwritten either way. If the confirmation itself did not finish, the strip says Confirmation did not finish: the prior version is still active and your edits are saved as a draft that can be confirmed again, and pressing the button again retries the same confirmation rather than making a second one. Discard my draft throws the pending edits away and shows the active version. Accepting a proposal under What your roles noticed lands that one field on the active version as its own revision and leaves any draft you have open alone.

Confirm this version stays off until you tick Confirm the reviewed draft, so a version becomes the one your team uses only when you mean it; hover the button to see why it is off. The box clears after each confirmation. Edit a new draft under the active context takes you to the first field of the proposed changes, and Rescan website takes you to the website field further down. A rescan of a confirmed brief writes proposals for you to review; it never changes the brief on its own. A newer brief and a confirmation that did not finish both show as a warning.

One brief for the whole workspace

Everyone in a workspace works from the same brief. When you add a teammate to your team, they read the brief you confirmed, their roles run on it, and anything they change lands in the same draft you see. Nobody keeps a private copy. If you work alone nothing changes: you are the owner of your own workspace and you confirm as before.

Any member can edit and press Save draft. Only a workspace owner can confirm, because confirming changes what every role in the workspace reads. For a member, Confirm this version stays off and says why when you hover it; pressing it from an older page saves the edits as the draft and tells you an owner confirms it. The strip above the form says whose draft is waiting. A member who accepts a card under What your roles noticed adds that change to the draft rather than to the active version. A member can discard a draft they saved last, and an owner can discard any draft.

Versions are numbered for the workspace, and each one names the person who confirmed it: v13 · Jordan. If a teammate confirms while you are editing, the strip says so by name, for example Jordan confirmed v13 while you were editing, keeps your draft, and offers the same field by field comparison and merged draft described above. The page only ever shows a teammate's name, taken from your own workspace; someone who has left is shown as a teammate, without a name.

When a team first moved to one brief, the team owner's brief became the workspace's. A teammate who had written their own brief keeps it, unchanged, and a copy of it sits in the workspace's version history marked with their name and never made active, so nothing anyone wrote was lost.

One business identity, across devices

The confirmed version is the server's, and each browser keeps a copy of it so a page can build a prompt without waiting on a round trip every time. That copy is stamped with the account and the version that filled it, and every page checks the stamp before it generates anything. Confirm a new version on your laptop and your phone picks it up on its next page; a field you removed from the brief is removed from the copy rather than left behind. A value you typed over the brief in Settings is kept, and the page says which fields you are overriding. Signing into a different account clears the previous company's copy before the new one is applied.

If that check cannot reach the server, a page refuses to generate and offers a retry instead of writing with a version of your business that may be out of date. Your saved targeting in Settings still wins over the brief's customer profile, and where the two disagree the page names the fields rather than leaving you with two answers.

Confirming fills Brand & Company

The brief is the source for Settings, Brand & Company. Confirming writes your company name, website, tagline, products, ICP and positioning into that tab, so every prompt a page builds carries the same company the scheduled roles read. Your elevator pitch, brand colours and logo stay with the tab's own AI Generate. If the tab already held something you typed, it keeps it; the tab's Fill from your business brief button applies the brief over it when you want that.

Confirming also gives the morning queue and the prospector an ICP profile when you have none yet: the industries, buyer titles and triggers from the brief. When there is no profile row at all, those surfaces read the confirmed brief's ICP directly. Approving a market profile in the launch writes its buyer titles into the profile when the row is empty or still the platform's own seed. A profile you edited by hand in Settings is left alone.

The guided flow applies the same Brand & Company keys after you confirm, and on any page where no company name is stored the shared AI client pulls them from your confirmed brief once per session, filling only empty fields.

The scheduled roles that draft, watch job boards and discover companies read only a confirmed brief, and stamp its version on what they write, so a draft can be traced to the brief that shaped it.

What your roles do with it

Your confirmed dossier is added to each role’s instructions as context: who you are, what you sell, who you sell to, and who you come up against. A role uses it to judge what matters, whose problem to lead with, and which accounts are on-strategy.

It is context, not evidence. Your ICP saying you sell to fintech CFOs does not tell a role that a particular CFO wants a call, and a role will not treat it that way. Where your dossier is silent, the role says so rather than filling the gap, which is the same rule your cited library already runs under. Sections you have left empty are left out entirely rather than sent as “unknown”.

If you have not built a dossier yet, nothing is added and your roles run exactly as they did before.

What your roles noticed

Once a day we read your own ledger: which accounts produced a reply, a meeting or a stage change over the past week. Where something has happened three or more times that your record does not reflect, you get a question at the top of Tell your roles about the business and in your line.

It proposes. It never edits your record. An account we noticed is held apart until you accept it, and until then no role sees it. Accepting adds it, with the observation as its reason. Declining closes the question and keeps what you wrote.

Three kinds of question. Accounts worth adding to your target list, and industries worth adding to your ICP: an industry needs three separate accounts behind it, not three wins at one company, because one loud customer is not a segment. If a night turns up nothing, you hear nothing.

The third asks whether a segment you named is still worth keeping, and it is the only one that removes something. Most ICP documents have not been touched since the deck they came from, and every role reads yours literally, so a segment still in it a year later steers drafts at a buyer you stopped winning. It needs all three of: your ICP untouched for ninety days, three or more accounts in that segment reaching your line without a single reply, meeting or stage change, and at least one industry that did produce outcomes and is not in your ICP.

The third condition is the one that matters. Without it, a founder in month one who has tried everything once and won nowhere would be told their ICP is stale, which is wrong: that is a motion not found yet, not a document that has drifted. Accepting drops those industries and leaves your titles and triggers alone. Declining keeps everything.

Corrections from your drafts land here too. When someone sends an edited draft or skips one under Activity, Drafts ready to send, the card asks what was off. A note becomes a line on one of four cards, one per kind: facts to correct, voice rules, decisions for the log, and phrases to ban. A card holds up to five lines and their titles start with Approve. Make these rules adds every line on the card as a rule; it does not change your brief. From the next run, every drafting agent reads the rules above your business context, the nine roles and the morning drafts alike, and a morning draft that uses a banned phrase is dropped. No thanks drops the lines, and anyone can raise one again from a later draft. Rules apply whether or not your brief is confirmed, because each one is already a person's decision. These cards are decided on this page only, where their lines are listed: Decide on Home brings you here and changes nothing by itself.

Rules your drafting agents follow lists every accepted rule with its kind. Retire stops one applying from the next run; the record of it is kept. A rule governs the wording and facts of drafts only. It never changes who is contacted, which steps need your approval, or which tools an agent uses. A banned phrase matches whole words, so banning "cat" does not touch "category", and a phrase found in more than a third of your recent drafts is refused when it is raised, because accepting it would drop most of them.

Completeness meter & consumed-by

At the top of /onboarding/dossier a completeness meter scores your saved dossier across its eight sections, company identity, products, founder/user, voice samples, ICP, go-to-market plan, target accounts, and competitor intel, and names the next one worth filling (for example, "4 of 8 sections filled, add voice samples to sharpen every product"). Voice samples matter most: they're what each writing product clones your tone from.

Consumed by shows which products have actually read your context. Core appears once your session pulls config (a verify_session or pull_config call), and each product appears with when it last synced brain state against your dossier, so you can see it flowing, not just saved. Until something reads it, it honestly says "not yet consumed by any product." A compact version of the same card rides along on your Activity page.

The meter reflects your last saved or confirmed dossier, confirm your edits, then refresh to watch it move.