An active recall client follow-up generator is retrieval, not recognition.

Active recall is the cognitive-science name for retrieving information from memory rather than recognizing it from a prompt or list. The classic example is studying with flashcards, where the act of pulling the answer out of your head strengthens retention more than re-reading the same page does. Applied to consulting follow-ups, active recall means writing the email by retrieving the actual commitments your client made on Tuesday and the actual phrasing you used the last twelve times you replied at this stage, instead of recognizing the deal stage and applying a generic template. The retrieved version is harder to fake and impossible to confuse with a sequencer email, because every concrete fact in the body came from a specific source you can point at.

No. CRM sequencers like HubSpot Sequences, Outreach, and Salesloft are recognition tools. They recognize the deal stage and the touch number, then they apply a template that was written once by whoever set up the sequence. The body is locked the day the sequence is built and the only personalization comes from merge fields. An active-recall follow-up generator does the opposite: every body composition starts from zero and is built up by retrieving specific facts from the call transcript and specific phrasing from your sent mail. The cadence rules can stay simple (send after call, nudge at day 7) but the content is freshly retrieved every time.

The Clone Computer Agent layer (described in src/components/architecture.tsx around line 18) opens the transcript page in your browser, the same way you would. It reads the page on screen. The Planner layer classifies sentences into commitments, decisions, risks, and small talk. A commitment is a sentence where one party said they will do something by a date or asked the other party to. Those are the lines that get quoted verbatim in the follow-up draft. You can see exactly which lines were retrieved because the draft review panel surfaces the timestamp ranges they came from. Nothing is paraphrased. Nothing is invented.

The mechanism is shown verbatim on the live product page at src/components/how-it-works.tsx step 03. Clone reads your sent-mail folder and observes patterns across your last twelve replies at each follow-up stage. The example in the source file is exactly this: 'Noticed a pattern in your last 12 kickoff emails: you start with a personal line, you always attach the SOW PDF, you cc your assistant when the contract is above $10K. Should I apply this template going forward?'. You confirm once. The pattern is saved to the Memory layer and reused. This is retrieval, not generation. Your phrasing is pulled from your outbox, not invented by a model that thinks it knows how you write.

Twelve is enough to retrieve the structural patterns: your opener, whether you bullet commitments back, your sign-off, your em-dash-versus-comma habit, your salutation choice. It is also enough to retrieve the conditional rules you actually follow without thinking, such as ccing an assistant when a contract crosses $10K. Below that, the pattern is noisy and Clone will surface it as a candidate rather than a rule. Above that, returns flatten out fast. Twelve is the working number from the source file (src/components/how-it-works.tsx line 45).

The existing guide at /t/consulting-follow-up-automation is about the same product but framed for the search of a consultant who already knows the territory. This page reframes the same retrieval mechanism through the cognitive-science lens. If you arrived here from a study or knowledge-work community where active recall is the working term, you are in the right place. The mechanics are the same: transcript context plus voice, both retrieved, never recognized.

Yes. The Computer Agent layer reads the transcript page on screen rather than going through an API. There is no paid-tier integration gate. If you can open the transcript in a browser tab, Clone can read it. This applies to tl;dv free, Fireflies free, Otter, Granola, and native Zoom cloud transcripts. Most solo consultants use the free tier of one of these, and Clone works against that tier without any upgrade.

No. The transcript stays in the transcript tool. The sent-mail patterns Clone observes are stored in the Memory layer on your disk, in a plain text file. The draft lands in Gmail. This is the first architectural principle on the live product page (src/components/architecture.tsx lines 46 to 49): Clone operates your desktop apps from your desktop, and client files, emails, contracts, and transcripts never leave your computer. The Planner layer may call a model to interpret your English instruction, but the attached client artifacts are not sent along with it.

Solo plan is $49 per month, with a 14-day free trial and no credit card required at signup. Boutique plan is $129 per seat per month for teams of two to ten. The transcript tool is whatever you already use, and the free tier of tl;dv or Fireflies is enough on its own. Compare against a virtual assistant who could draft your follow-ups under your supervision at $3,000 to $6,000 a month, or against CRM sequencer tiers that range from $20 a seat per month for HubSpot Sales Hub Starter up to $100 a seat per month or more for Outreach and Salesloft.

Every draft lands in Gmail as a draft, not as a sent message, on the first pass and on review_first stages. You see the draft. You fix the line that was wrong. The fix goes back into the Memory layer as a voice-pattern update so the next draft at that stage retrieves from a slightly more accurate sample of your writing. This is the 'always reviewable' principle (src/components/architecture.tsx lines 60 to 63): every action Clone takes is logged and reversible. For most consultants, the most common correction in week one is a phrasing edit, and by week three those corrections become rare.

Yes. Clone treats the CRM as a source of truth for the engagement stage, not as a required piece of software. Plenty of solo consultants run a single Google Sheet as their CRM and Clone works against the Sheet the same way it works against HubSpot or Pipedrive. The retrieval still pulls from the call transcript and your sent mail either way.