What shipped, and when.

Your conversations are waiting tomorrow, and a presentation is redrafted when its page moves on

• A conversation is where you left it after a refresh. What you ask Liffey is saved as you go, so a closed tab, a reload or a return the next morning finds the thread under Resume a conversation, with its title, how long it ran and when you last spoke. The newest thirty are listed, and opening one brings its messages back.
• How long a conversation is kept follows your plan. Operating keeps 30 days and Extended keeps 90. Where Liffey runs on your own infrastructure, your conversations are kept for as long as you keep them, under your own policy.
• Contact details are taken out before a conversation is stored. Email addresses, phone numbers and account numbers are stripped from the text and the title on the way to storage, in every workspace, the same treatment a search query already gets.
• A presentation is redrafted for your review when the page it presents moves on. Every hour Liffey looks for pages whose writing has changed and settled, then drafts a fresh presentation into your review queue. The deck your colleagues and clients are reading stays exactly as it is until you approve the new one, so the slides move when you say so.
• A presentation written after its page reads as current. Where the presentation is the newer of the two, your workspace says so, and the note that the writing has moved ahead of the slides appears when the words themselves have changed.

Asking for a brain takes one field, and the company is confirmed before a page is read

The sign-up at liffey.app/start asks for a work email. The address names the company, so Liffey confirms which website it is about to read, then builds while you watch it run.

• A work email is the whole of the form. The address gives Liffey the company and the website together, so the build has what it needs from the moment you type it. Your name and phone are offered later, while the build runs, for anyone who wants to add them.
• The company appears under the email box before anything is sent. Leaving the field, Liffey answers which website it is about to read, so a mistyped domain is caught before it costs you a wait. A personal address gets its answer in the same place, ahead of any code.
• The wait page shows a build that is alive. The Liffey mark breathes and sends a ring outward for as long as the build is waiting for approval, queued or running, and it rests once the build has settled. A machine set to reduce motion gets a still mark.
• A website that turns automated readers away is still read. Some company sites are guarded by a service that answers a laptop and refuses a server. Liffey re-reads those pages through a hosted browser the moment a site declines it, on the crawl as well as the first read, so a guarded website builds from its own pages. Every read of your site now goes through the same reader and identifies itself the same way.
• A build stops when the website gives it nothing to read. Where the pages cannot be reached, the build reports that to a commercial architect and waits for a person, so every page of your workspace is written from something Liffey read.
• Your workspace is drafted in batches. The Foundation writing and the presentation behind each page are drafted several at a time on every pass, so a workspace finishes in fewer passes. A page that fails is simply still missing on the next pass, and the finished work beside it is kept.
• A declined request ends with a call you can book. The mail that turns a request down offers a time at liffey.ai/book-a-call.
• A stat card reads as a number, then its caption. In every presentation, including the ones already written, the figure and its caption take their own lines, with space above the row.
• Your workspace is named beside a square brand mark. Where a company's logo is a square mark, the navigation names the workspace next to it. A wide wordmark already says the company's name and stands on its own.

Your meetings reach your brain from the notetaker that was in the call

Granola and Pocket join Otter as accounts you can connect, so the notes and transcripts your team already captures index themselves. A meeting the video platform wrote up itself goes in as the caption file it produced.

• Connect Granola with a sign-in. Recent meeting notes and transcripts from the calls Granola took notes in are indexed into your brain. Granola names what its server offers in its own way, so Test connection asks the server what it can do and confirms Liffey found something that lists your meetings before the source reports as connected.
• Connect Pocket with a key from your own Pocket settings. Recent conversations are indexed with their full transcripts, newest first, on every Pocket plan.
• A caption file from a video call reads as conversation. The transcript an organiser downloads from a video meeting is a caption file, and most of what is in it is cue numbers and clock timings. Liffey turns those cues back into what somebody said, attributed to the speaker where the platform recorded one and joined into paragraphs a reader can follow, so the meeting is what reaches your brain. It goes in anywhere a file does: an upload, a paste, your own AI assistant, and the transcript box on a meeting.
• Liffey reads these accounts and leaves them as your team left them. Each connection asks for the access that reads your meetings and indexes them.

Every page waiting on evidence is named before your workspace opens

Liffey marks the spot where it cannot source a quote or a figure, which keeps invention out of your first draft. Those marks now travel with the workspace to the commercial architect who releases it, so the pages waiting on evidence are the ones they open first.

• Your first draft shows where a connected source would change the answer. Each place marked for evidence Liffey could not find is counted page by page and reported with the finished workspace: in the mail that says it is built, in the release queue, and in the confirmation at the moment of release. The components carrying most of them are named, so a commercial architect reads those pages first and decides what to do about them before you open the workspace.
• The editorial standard follows your workspace as it grows. Every change to a component is measured as it is written. Edits made in the app, publishing, writes from the assistant, the overnight content scan and the nightly refresh all pass through one point, and the result is recorded with the component there, so the standard that governs your first build governs your workspace afterwards.
• The score counts the words a reader meets. The measurement reads the copy on the page, so the number describes the writing your market will see.
• Headlines break where a person would break them. A slide headline sits on the measure display type is set for, which keeps a phrase whole and keeps the last word off a line of its own. Decks already in your workspace pick this up when they open, and every deck and document written from here carries it.

One editorial standard behind every word Liffey writes for you

The first draft of a Strategic Foundation is writing somebody will put in front of their own market. It is now written to the same editorial standard Liffey applies to its own copy, and measured against that standard before a commercial architect releases the workspace.

• Every word the platform drafts for you follows one editorial standard. Your Strategic Foundation components, the writing you get when you ask the assistant to draft something, and the slides behind each component are all written to the same rules: affirmative claims, plain sentences, and the verb that names what actually happens. The rules govern how a sentence works, so your own language and your own positioning stay yours.
• Your workspace is measured against that standard before you open it. Each build scores the copy it produced, page by page and deck by deck, and reports the result to the commercial architect who reads your workspace before release, so the pages worth a second look are the ones they open first. Each workspace is measured against its own first build, which makes your writing the benchmark, and the benchmark tightens from there.
• Your home page points you at work that is genuinely waiting. A new workspace opens with its Strategic Foundation already in draft, so the prompt you meet names the overnight scan and the decision it needs from you.
• A build that stumbles finishes on its own. A step that fails is tried again from the point it reached, keeping every component already written, and a commercial architect is told every time it happens. The mail announcing your workspace names anything that needed a second attempt on the way, so a clean report means a clean build.
• The mail that opens your workspace says what it was built from. It names your public record as the ground under the first draft, then what sharpens it: the conversations you are having with your customers first, since your buyers' own language lives there, then your CRM, your analytics and your written strategy.
• Coming back to liffey.app/start keeps your place. Open the terms in a new tab, switch apps, or return with the back button, and the sign-up resumes at the code we emailed you, with the details you typed already in the form.

The revenue tables open in a spreadsheet, and a pasted link shows where it leads

A figure is worth more when it travels. This release takes the revenue report into the tools finance already works in, and makes a workspace link legible the moment somebody pastes it into a chat.

• Export the revenue tables to a spreadsheet. The month accounts table and the year grid each carry an Export control, and the file arrives with the same columns, the same rows and the same order as the screen it came from. Money goes out as a plain number at full precision so a spreadsheet can total it, with the currency named once in the header. The file names itself in words, giving the business, the report and the month, so it still reads clearly in a downloads folder a year later. Accented names, a comma inside a customer name and text a spreadsheet would otherwise run as a formula all arrive intact.
• A link into your workspace previews as its destination. Paste a board link into Slack, Teams, LinkedIn or a message thread and the card names where the link goes and carries your own mark, drawn as a picture in your accent colour. A preview is fetched by a crawler with no session, so what travels is your workspace name and the name of the board, and the content behind the link stays behind sign in.
• The pipeline board draws the stages your CRM uses today. Where a team has customised its stages, the board follows that customisation, and a deal recorded under a stage the CRM has since retired keeps its history.
• The year revenue table totals the accounts as well as the money. The total column carries the account count for the year, counting every account that billed at any point in it or is projected to, so an account that churned in September still counts as one the business served this year.
• Scaled figures read to one decimal place everywhere, so 48.2 K and 1.3 M read the same way on every board, in every summary and in the overnight scans.
• The revenue year view is an address you can send. A given year and a given month each have their own link, and the address bar keeps up as you move around the report, so a view you reached by clicking is a view you can paste to a colleague. Month labels on the chart and in the table header open that month.
• The trust page states the retention each record actually runs on. The audit log is described as it behaves: append-only, held for the life of the workspace, and removed when the workspace closes. Product usage records, the page activity an owner's view of team activity is drawn from, are named in the data Liffey holds and carry their own retention row, deleted automatically after 90 days.

Every commercial number in one place, and each one says where it moved

The pipeline, the customer health board and the revenue report ask different questions of the same business. The Commercial Scorecard now opens all of them, and every figure on it is computed by the same library the board behind it uses, so a summary and its deep dive agree by construction.

• The Scorecard reads as the commercial motion in order. The funnel leads: how one stage becomes the next. Beneath it sit acquisition, pipeline, customer health and revenue, each with its headline figures and a way through to the board behind it. A block still waiting on a source shows you where to connect one.
• Every figure carries its change on last month, as a percentage. Counts and money sit side by side on that panel, and +66% beside +31% compares in a way two different units cannot. Acquisition reads the rolling 30 days against the 30 days before it, so the top of the page is as current on the 2nd of a month as on the 28th.
• The pipeline says what the month lands at. Open and weighted stay as they were, and two figures join them: expected, the weighted value of open deals dated this month, and forecast, closed won plus expected. Open, weighted and forecast each carry their movement on last month.
• A month is measured against its closing position. A CRM answers what is true now, so month on month reads a snapshot, and that snapshot is now taken on the last day of every month for every workspace whatever day its review runs. The date arithmetic behind it is fixture-tested against leap years, short months, the December rollover and the exact hour the job fires.
• Money follows your CRM's own currency. A business billing in sterling reads in sterling wherever its pipeline is summarised, including inside the overnight scans that reason over those figures.
• One house rule for every number in the app. Anything a thousand or over is scaled and carries its unit to two decimal places: 23.45 K, 1.34 M.
• The customer health meeting is an occasion with a date, and it keeps a ledger. Every rating change and every status written is recorded beside the rating it replaced, so months later the trail reads as "green to amber" and answers why an account moved, and when.
• Your health meeting recording rates the accounts it discussed. The rating comes from the language people actually use, "all good", "at risk", "unhappy customer", and the status is written in the team's own words beside it as the evidence for it. The pass touches health and status, leaving revenue to the model you authored, and an account nobody mentioned keeps the rating it has.
• Anything a machine wrote can be corrected by hand. A deal note on the pipeline board is now editable by owners and admins, with the earlier wording kept in the audit record, which is what makes writing straight through from a recording safe. Health ratings and statuses are editable from the row.
• A note carries over per deal, dated. Every deal keeps the last thing said about it for the whole of the following review, greyed and stamped with the day it was said, so the context is there at the moment the meeting is running.
• Both boards are operated the same way. One header shape across pipeline, customer health and revenue, the same pencil beside the review date, and the same transcript control taking Otter, a paste or a file upload. "Last review" now reads the ledger, so it is the last time somebody genuinely changed something. Wherever you are in the app, the sidebar shows you where that is.

Your revenue plan and the health of every account, on one set of numbers

The same accounts, asked two different questions. The revenue report leads with the money and carries health as a rating beside it. The customer health board leads with the account and prices what a rating is worth, because a health score nobody can put a figure against is a health score nobody acts on.

• Your revenue plan reads the way finance already writes one. Bring in a sheet with the months across the top and a row per account, or the long shape, and either is read as it stands. Spellings that differ for one customer are folded together so a real concentration stays visible, and every fold is reported in the preview for you to accept.
• Revenue you hold and revenue you expect stay apart. A line rated green, amber or red is money the business has; a prospect is money it hopes for. Every figure arrives in both forms, so the gap between them sits on the page as the exposure it actually is.
• The prospects come from your CRM, so the block you kept by hand is done for you. Deals sitting in a stage rated 60% or better are read from your pipeline snapshot and spread across the twelve months from their close date. Your CRM stays the authority on its own deals, and a customer already in your own model is left exactly as you wrote it, so your sheet is the override. The page carries the date of the snapshot it read.
• The month view opens on the month you are standing in. Arrows step through the months and the dropdown jumps. The table carries what your working sheet carries, health, status and renewal, and adds each customer's change on last month, so a service line that ended is visible at a glance.
• A year on one canvas, money and accounts together. Confirmed and expected draw as their own lines with last year behind them. Hover any month for the real figures, and tap either series off to read the other on its own.
• Rate an account at the moment you are looking at its number. Health, status and renewal are editable from the row itself, and the rating applies to the whole customer, whichever service line you happened to open.
• The health board opens worst first. Revenue at risk is the join between the two pages: what this month is worth and how it moved, how many customers you have and how that moved, then the same month and the same customers split into green, amber and red, so the bands add back to the line above them. Declines against last month and renewals falling inside three months sit under the summary, since a renewal is the moment a rating gets tested.
• A status carries the last thing anyone said about the customer, and the date they said it. A weekly meeting reaches some accounts and not others, so the ones discussed read at full strength and the rest keep their earlier note, greyed and dated. Only a status that genuinely changed moves the stamp, so the page stays honest about which accounts were really talked about.
• A customer billing for the first time says so beside their name, worked out from whether any earlier month carried revenue, so a customer who pauses and comes back keeps the history they already have.
• Each meeting keeps its own day. The pipeline review, the customer health meeting and the commercial leadership meeting are separate rituals with separate people, and each now carries the day it actually runs on. A team already running its cadence keeps the day it chose.

The estate is proved every morning, and a workspace is drafted from your whole site

Two halves of the same discipline. The workspace that builds itself once a request is approved now reads the whole of your public record and arrives under the name you trade as, and every live workspace is proved to be working before anyone opens it.

• Every morning, every live workspace is proved to be working. A suite runs at 06:00 and asks a positive question of each one: the sign-in page renders, it carries your name, and it offers you a way in. It does the same for the pages a visitor meets before signing up, checks that every brand asset a workspace names is really there, and reports any build that has stopped moving. It writes only when something needs attention, so a mail from it is a mail worth opening.
• The morning check is itself checked. A page that has failed can still be served as though it were healthy, which is how a broken workspace passed inspection more than once. Each check reads the page itself rather than trusting the response, and a canary proves every morning that the judgement can still tell a failed page from a good one. If that ever slips it is the first thing reported, because every other pass rests on it.
• Your first draft is written from across your website. The build finds your pages the way a machine is meant to, through your llms.txt and then your sitemap, and reads across the site to the same page limit as a crawl you run yourself from Web Links. It reports how much it read against how much it found, so the depth of the grounding behind your Strategic Foundation is on the record from the start.
• Your workspace carries your company name. The name is taken from your site and resolved to the company itself, with your web address standing behind it, so the workspace opens under the name your clients know you by.
• A build that pauses is picked back up within minutes. A standing check runs through the day and resumes any build still in flight, carrying on from the step it had reached. Every step asks what is still to be written, so a build can resume at any point and arrive complete.
• Your own brand on the first page anyone opens. A workspace running on its own dedicated instance carries its name and its mark on the sign in page, so the address you hand your team opens as yours.
• An address is answered honestly, and gives nothing away. A platform administrator visiting an address where a workspace has been removed is told exactly that, which is what makes the answer useful to them. Everyone else receives the answer they have always received, so an address stays a poor way to learn which workspaces exist.

Your workspace builds itself, and a person reads it before you do

Asking for a brain at liffey.app/start now carries all the way through to the mail that tells you it is open.

• An approved request builds itself, and keeps its place. Your workspace is drafted from your own website as soon as a commercial architect approves the request: your brand, then your Strategic Foundation, with a presentation behind every written component. The build works a step at a time and records each one as it lands, so it carries on from the point it reached, and a build that stalls is picked up again and finished.
• A commercial architect reads the finished workspace before you receive it. A build that completes goes to a person for review. Your account is created at release, and release is what sends your invitation, so the address opens to you once somebody has read what was written for you.
• You are told the moment it is live. The mail gives you your web address and asks you to sign in with the email you signed up with and a code we send you. It says plainly that what is waiting is a first draft written from your public record: a starting point, yours to change wherever it reads differently from how you would say it.
• The review that happens before release is time limited and recorded. The architect who opens your workspace to read it does so on an elevation lasting two hours, written into the same audit trail as every other administrative access.

A workspace arrives with its slides, wearing its own brand

A workspace built from your website now arrives the way a commercial architect would hand it over: the written strategy with the presentations beside it, in your own brand from the first page you open.

• Every written component arrives with its presentation. Slides are drafted with the writing as the workspace is built, on the same house scaffold and the same export contract as a deck you ask for in conversation, so each piece of thinking opens with both halves in place. A longer component is drafted again, tighter, so it keeps its slides too.
• Your brand comes off your own website. The build takes the logo your site actually uses, the reversed version of it for dark surfaces such as the navigation and deck covers, and your square mark for the browser tab and the collapsed navigation rail. Each is stored in its own right, so the mark that appears in a place is the one designed for it, and the web address the build worked from is kept on your Brand page for whenever the artwork changes.
• The patterns a synced call surfaces are worked one at a time, above your board. They sit in a strip of their own between the page header and the pipeline board: read one, share it with the team, and the strip moves you to the next one still waiting on your decision. It closes by naming what you shared, with a link through to Insights and Tips, and it says up front that these patterns last for the visit, so you know to act on them while you are on the page.
• Connecting a tool brings you back to your own workspace. Whichever database instance your workspace runs on, authorising a tool at the provider returns you to your connected sources page with the connection in place.

The road from sign-up to a workspace that opens

Between asking for a brain and opening one sit a handful of automated checks on the company behind the request. A mistyped web address should cost a correction, and most of the day went on making that true. The trust page, which a procurement team reads before any of this begins, gained the two disclosures it was still carrying in outline.

• A held sign-up can be corrected and sent again. Where the check on your company web address holds an attempt, liffey.app/start offers "Check the details and try again" with what you entered already in the form, so a single wrong letter in a domain costs a correction. A repeat attempt always goes to a commercial architect for a decision, so trying again reaches a person.
• Confirming your email tells somebody you are on your way. Verification now alerts a commercial architect as it happens, and names which attempt this is, because somebody who comes back is the signal worth acting on. The alert is raised after your own confirmation has completed, so it stays clear of the thing you came to do.
• A held build is corrected in place, and every decision leaves a record. Liffey's architects work from a queue that leads with whoever is waiting on a decision. The web address can be corrected there: the check runs again against the corrected address, and the build is queued the moment it agrees. Approving a build as it stands is still a click, under a label that now says exactly that. The audit trail records whether a person changed the details or accepted them as submitted, who did it, and from which door.
• A new workspace opens on every page from its first minute. A workspace built from your website carries your logo straight away, and wears the Liffey mark as its browser icon until a square one of your own is in place. Brand detail is decoration, and the pages stand on their own whatever it holds.
• The chat panel keeps the width you gave it, everywhere you work. Ask Liffey and the component workspace now share one right-hand panel: drag its edge and that width follows you from a component page to settings and back. On a narrow window it lays over the page, keeping the work beneath it at full width, and the main and settings navigation fold to suit the room they have.
• The trust page describes a stored call review, and names the sources you can connect. The table of what Liffey holds now sets out a review in full: a summary, each criterion with the transcript line that earned its verdict, one recommendation, and what the customer said that the business should know. It states that personal identifiers are stripped before the transcript reaches the model or the database, whatever your workspace redaction setting says, that the review is what is kept, that reviews age out on your retention window, and that a member can delete their own at any moment. The sub-processor table now names Notion, HubSpot, Zoho CRM, Google Analytics, Google Search Console, LinkedIn Ads, Otter, Canva and Featurebase alongside those already listed.

Every call your team has, read against your own architecture

A call happens at a point in a relationship, and your commercial architecture already describes what good looks like at that point. So call review reads the page your team wrote. A discovery call is read against the problems you say you solve and the people you say you serve; an onboarding call against your onboarding. The standard is yours, and it now reaches every member of the team rather than the sales side alone.

• Name the kind of call, and your own thinking becomes the yardstick. Discovery, deal conversation, handover, onboarding, customer check-in, renewal or expansion, partner, and voice of customer interview. Liffey suggests the kind from the recording title, and you can correct it, because meetings are titled by whoever booked them.
• You are told what will read the call before you spend the time. Choose a kind and the line beneath names the part of your brain it will be graded against, so the yardstick is visible up front rather than inferred afterwards from the feedback.
• Where your brain has yet to describe that kind of call, Liffey says so and reads openly. What went well, what could have gone better, what emerged. Your own thinking is the only standard it will apply, so the card names the gap as a piece of work your commercial architect can pick up, and calls like it are measured the moment that page exists.
• Bring recordings in from your own Otter account. Each member connects their own and sees only what they recorded, because a list of who somebody met and when belongs to them.
• Each call is read in full and appears as it lands, with a running count as the batch works through, so a wait that is genuinely a wait looks like one.
• What a customer said now travels to your team. Every call is read for what a customer said about their problem, a competitor or your price, not only the research interviews, because the most useful thing a customer says is usually said in passing. Each insight carries the words they actually used, and one that cannot be traced to a quote is held back rather than shown.
• One button puts it in front of the team. An insight goes to the team wall, and where the customer spoke with real warmth or real frustration it also reaches the owner review queue, so that queue stays short enough to be read.
• The most useful line on the card is now something you can act on. "Do this next time" carries the single highest-value change alongside the transcript line that earned it, and shares to the team in one click, so a recommendation arrives with its evidence attached.

Confidentiality in the terms, and every way in set out in full

A client who hands over their commercial strategy is trusting the people as much as the platform. The terms of service now carry that commitment in writing, and the trust page describes every way your brain is reached, so a procurement or security review can read the current position in one sitting.

• Confidentiality is a section of the terms. Mutual, covering everything held in your workspace: your strategy, positioning, pricing, customer and pipeline information, the material you connect and the decisions your team records. It runs for the term of your use and for five years after it, and anything that qualifies as a trade secret stays protected for as long as it remains one.
• Everyone who works on your account is bound in writing first. Written confidentiality obligations, at least as protective as the ones in your terms, are a condition of engagement for every person on your account, whether employed by us or engaged as a contracted architect. Those obligations are in place before access is granted, and we stay responsible to you for them.
• Access to production is granted for a stated reason and a limited time, with every grant recorded in the audit log and alerted.
• The terms travel with you. Sign-up states plainly that continuing is agreement and names the confidentiality commitment inside it, both sign-in pages carry the terms beside the privacy statement, and every page of your workspace now reaches both from a footer, so a member invited into your brain can read what they are bound by from wherever they are standing.
• The trust page names each way your brain can be reached, the app, an AI tool your team already uses, Slack, and the web API and command line, with what each may do, what holds before an answer is given, and the fact that each is off until an owner turns it on. Every read through any of them is recorded in your audit log with the person, the way in and the time.
• Programmatic access states its capabilities in full. A token's reading and its writing are set separately and each starts at the safer setting, the holder's role is re-checked on every call so a change of role applies to the very next one, and revoking a token takes effect immediately.

Hand your existing material to your brain from your own AI

The collateral a business already has, its sales decks, case studies, pricing sheets, onboarding guides and call notes, can now go into your brain straight from the AI tool you work in, and the page and its deck can be written from that material in the same sitting.

• Your connected AI can add a document. Claude, ChatGPT, Copilot, Gemini or Mistral hands Liffey the text, and it is searchable from that moment, on every surface that reads your brain.
• It arrives exactly as an upload does. Your redaction setting, indexing and retention window apply the same way, so where a document came in from changes nothing about how it is held. It appears on its own line in your sources, ready to be archived or removed like any other.
• Working through a folder is safe to repeat. A document sent again under the same name replaces your earlier copy of it inside your own brain, so a job finished over two sittings leaves a single pricing sheet, not four that all match every search.
• Adding material is separate from changing your strategy. Raw material is something your team can read and weigh, so it goes in on a key that may write; rewriting what your team has agreed keeps the owner review step it has always had.
• A page can gain a presentation. Ask a connected tool for a deck on a component that has only words, and Liffey hands it one of your existing decks to take the house style from, so the new one matches the brand your team already set rather than an invented look.

Your written strategy became searchable

Liffey has always searched everything you connect: your site, your documents, your CRM, your calls. It now searches your written commercial architecture the same way, so a question about your pricing model reaches the component where your team decided it.

• Semantic search over your components. Ask in your own words. "How much do we charge" now finds the page that talks about fee tiers and annual commitment, whether or not it uses the word charge.
• Two ways of searching, working together. Exact phrases, product names and figures are matched literally. Meaning is matched semantically. Each gets a place in every set of results.
• Every surface benefits at once. The same improvement reaches the in-app chat, Slack, your connected AI tools and the web API, because they share one retrieval path.
• It stays current on its own. Your strategy is re-indexed whenever it changes, so an answer reflects the page as it reads today.

One page to connect Liffey to the tool you already use

• Pick your tool, then follow its steps. Claude, Slack, ChatGPT, Gemini, Microsoft 365 Copilot, Mistral, Claude Code, or the terminal and web API.
• Claude and Slack need no key. Paste one address, approve what it may do, and Liffey keeps the connection alive in the background.
• Connected means it really called us. A green light appears the first time a tool genuinely reaches your brain, and never merely because a key was created. An assistant can answer from its own general knowledge and sound the same, so the product holds itself to the stricter test.
• The same row sits under the dashboard question, so the capability is visible where people already work rather than only inside settings.

Ask your brain from Slack

• /liffey in any channel, plus @mentions and direct messages with Liffey.
• Answers are private to the person who asked, so sharing to a channel stays a deliberate act by somebody who has read it first.
• Only your team gets answers. A Slack workspace holds contractors and guests, so every request is matched to your brain's own access list.
• Sources are linked, so any claim can be opened and checked.
• Read only. Nothing in Slack changes your brain.

Work on your brain from your own AI

• Connect your brain to Claude, ChatGPT, Gemini, Copilot or Mistral, and work on your commercial architecture from the tool you already have open.
• Sign in rather than paste a key. Liffey issues the connection itself using the current authorization standard, so approving a tool takes one click and access renews on its own.
• Three levels, chosen by you. Read, propose, or commit. A proposal lands in the owner review queue with a diff, exactly as an in-app suggestion does.
• Permission follows the person, not the key. Access is re-checked on every single call, so a change of role takes effect immediately.
• Presentations too. A connected tool can read and update a component's deck alongside its written page, so the two stay in step.
• Every read and write is recorded. npm run usage reports people, sign-ins, questions asked, pages read and changes made, over time, which is how an engagement shows what it is worth. Search text is stripped of personal data before it is stored.

The commercial architecture reorganised around the customer

The four pillars were rebalanced after a working session with a customer-success practitioner, so the post-sale half carries the same weight as the pre-sale half.

• Customer health joins the operating system beside pipeline review, because the health of your customers is an operating rhythm rather than a one-off exercise.
• A customer journey component, paired with a renamed sales journey, so both halves of the relationship are described in the same language.
• Customer metrics on the scorecard, so retention and expansion sit next to acquisition in one view.
• Team maturity folded into the operating system, where the work of building the team actually happens.

Commercial Pacing, the weekly leadership cadence

• A standing register. An opportunity stays open until it is closed, and every open one is automatically on the next meeting's list in priority order. Nothing to carry forward, nothing to reschedule when a week is skipped.
• Drag to prioritise, with keyboard and touch alternatives.
• Concluding a meeting emails each person their own actions, as a plain list ready to paste into whatever they already use.
• Upload the recording and Liffey writes it up. Discussion, actions, new opportunities and challenges to your strategy are all proposed as drafts. You edit and accept them; nothing is written by the recording itself.
• Resolving an item teaches the engine. The outcome and your reasoning are fed back, so the same evidence is not raised again under a new name.

Self-serve sign-up at liffey.app/start

• Five fields and an email code. Verifying the inbox is the sign-in.
• A stage gate before anything is built, so a workspace exists only once a build is approved.
• Request access when your company already has a brain, landing as a card on your owners' team page and as an email to each of them.

The patient insight engine

The engine stopped judging each night in isolation and started remembering.

• Signals build across nights. A pattern seen once is held; a pattern that keeps recurring, from more than one source, is surfaced for a decision.
• A weekly consolidation pass merges strands that are one pattern described two ways, which a single night structurally cannot notice.
• Corroboration before it reaches you. When something crosses the bar, Liffey sweeps your whole brain for evidence for and against, inside your approved sources only.
• Four answers to every finding: set aside, monitor, raise with the team, or lock it in.
• Locking it in writes your conclusion into the component, dated, attributed and with the conditions that hold it true. Those conditions go under watch, and Liffey raises a decision check when they break.
• Anyone can escalate. A challenge raised by a person is weighted more heavily than one the engine noticed on its own.

Advertising performance on the Commercial Scorecard

• Ad spend and return sit beside the rest of your numbers, so acquisition cost is read next to what it produced rather than in a separate tool.

The assistant can pull in a live page mid-conversation

• Ask about a page and Liffey can go and read it, through the same checked path the manual web-link flow uses.
• Anything it fetches becomes a real source in your knowledge base, tagged with its role: your own material, a competitor, or industry signal.

The review queue became one page you can act on

• Two clear sections, what the nightly scan raised and what your team proposed, with the same actions in both.
• Approve a suggestion and see the actual diff before it lands.

Retention windows

• Set how long personal-heavy material is kept. Call recordings, email and chat are distilled into a permanent period summary before the raw records are removed, so the knowledge stays and the personal detail does not.
• Your written brain, decisions and scorecard history are always kept.

Decision traces, learning loops and first-party analytics

• Conclusions recorded as decisions, dated, attributed, classed, with evidence attached and the conditions that hold them true.
• What we have learned accumulates against each component as challenges are accepted.
• Team activity for owners, first-party only, with no third-party analytics on any workspace page.

Zoho CRM

• Zoho joins HubSpot as a first-class CRM, feeding the same scorecard and the same pipeline board.
• Won and lost are classified from the data, never from the names your team gave the stages.

Personal data stripped at ingestion

• Turn it on and email addresses, phone numbers and bank details are replaced with typed placeholders before anything is stored or indexed. Names are kept, so the material still reads as itself.
• Applied at the two points everything passes through, so it covers every source rather than every connector having to remember.

Reconciled figures, and the review value loop

• The scorecard and the pipeline board agree, because both read the same underlying deals the same way.
• Every judgement is recorded as an event, which is what makes a value report possible later.

Dedicated database instances

• A client can have their own database rather than a shared one, with the same product on top and the same upgrades arriving at the same time.

Lead-magnet whitepapers, and decks into Canva

• A repeatable whitepaper pipeline, from seed to a finished asset in your hub.
• Export a component's presentation into Canva or PowerPoint and keep editing it there.

Editable exports, and the nightly scan splits in two

• Presentations export as real editable files, with client-safe filenames.
• Members see opportunities; owners see the challenges. The nightly scan routes each to the person whose job it is, with an optional digest by email.

Onboarding

• Brand set-up reads your site and fills in your colours and marks.
• Get started on any component, with the questions that component needs and a first draft offered from what you have already connected.
• Readiness signals show what a component still wants before it is useful.

Content Radar, and the Assets workspace

• A weekly thought-leadership engine that reads your strategy and your chosen industry sources, then proposes topics with attributed evidence and a live source link.
• Prepare a draft from any topic, anchored in your positioning rather than in general knowledge.
• The Assets workspace for building and editing presentations in your own brand.

Bring your own AI, two ways

• Route Liffey through your company's own governed model, whether that is Anthropic, Gemini, or any OpenAI-compatible gateway.
• Or connect your brain outward into the AI tools your team already uses.

Call recordings, Ask Liffey everywhere, weighted pipeline

• Connect your call recordings so what customers actually said is part of what Liffey reads.
• Ask Liffey from any page, in the context of what you are looking at.
• A weighted pipeline view, reading live from your CRM.

Connector metrics and the Commercial Scorecard

• Your connected systems feed a single scorecard, each metric shown against the prior period.
• A commercial funnel strip across visitors, leads, deals, closed and customers.

EU residency complete, and the trust page

• Database, compute, AI inference, embeddings and content store, all in Europe.
• A public trust page whose every claim is checked against the code before it ships.

The home dashboard, artifacts and the Knowledge Base

• One question at the top of the workspace, with the four pillars beneath it.
• Artifacts in chat, so a draft arrives as something you can work with.
• The Knowledge Base, showing everything Liffey can currently read.
• Roles and access control, invite-only, per workspace.

Insights, the brain and front-end split, and email into your brain

• Insights surfaced on the home page, opportunity-led.
• Two layers, kept apart on purpose. The written components are the human-verified truth; the knowledge base is everything you have connected. Committing to the brain proposes a full revised page for review rather than overwriting anything.
• An email address for every workspace. Forward something and it joins your knowledge base.
• One row per thing you add, with archive and purge on each.

Retrieval, the settings hub, and the command palette

• Retrieval that asks your question several ways before answering, so a phrasing you did not think of still finds the passage.
• A settings hub for sources, models, team, brand and billing.
• Command palette on ⌘K.