An AI clone for consultants is not an avatar. It is the thing that drafts your thirteenth kickoff in your voice.

Operational clone, not avatar. The product reads your last twelve kickoff emails on the Gmail thread, extracts three rules you follow without thinking, asks to save them as a named ritual, and then drafts the next kickoff inside your Gmail tab using those rules. The reference UI is in src/components/how-it-works.tsx step 03 on cl0ne.ai. The three example rules ('personal opening line, attach SOW PDF, cc assistant when contract is above $10K') are not marketing copy, they are the literal example output the product surfaces, hard-coded at lines 44 through 56 of the live file. 'Saved as default-kickoff' on line 56 is the operational definition of cloning a consultant in this product.

Avatars produce a video of your face. They are useful if your bottleneck is producing video, which a consultant's bottleneck rarely is. The Monday morning queue is invoices to draft, follow-ups to write, and CRM rows to log. An avatar does none of those things. The operational clone does all of them, by driving your existing Gmail, QuickBooks, and HubSpot sessions from rituals it learned from your real history.

Voice clones synthesize audio from a five-minute sample. They are excellent for podcast intros and explainer videos. They cannot draft a follow-up, push a CRM note, or queue a Tuesday 8am invoice run. Clone in this sense operates the apps; it does not generate audio.

Knowledge clones train a chatbot on your blog posts so it can answer questions in your style. They are useful as a public-facing FAQ on your website. They do not touch your private Gmail thread, your QuickBooks invoice draft, or your HubSpot pipeline. The operational clone in this product reads what is already in those apps and writes back to them.

Open src/components/how-it-works.tsx on cl0ne.ai. The full step 03 lives at lines 42 through 58. Line 56 contains the literal string 'Saved as \'default-kickoff\''. Lines 44 through 53 contain the three rules the pattern-mining UI surfaces verbatim. The principle that drives this behavior is in src/components/architecture.tsx at lines 50 through 54: 'Clone observes how you draft emails, format proposals, and close engagements, then mirrors that style. It's your working habits scaled, not a generic template library.'

No, the default is review-then-approve. The drafted kickoff lands in your Gmail drafts, queued for one click. You can opt into auto-send for a specific ritual once you trust it (the Monday 8am invoicing ritual is the most common candidate), but the out-of-the-box behavior is 'draft and queue, never send', explicit in the step 02 sequence at src/components/how-it-works.tsx lines 22 through 40.

Anything that has a stable pattern in your last N runs. Invoicing: the last 6 invoices give it your rounding rule, your line-item phrasing, your due-date convention. Follow-ups: the last 10 post-call emails give it your subject-line shape, your CTA phrasing, your action-item formatting. Status updates: the last 4 Friday emails give it your three-section template (what was done, what is next, what you need from the client). Each ritual is named, lives as a markdown file in ~/.clone/memory/rituals/, and is editable by hand.

Your Mac. principle 1 in src/components/architecture.tsx at line 46 is the commitment: 'Client files, emails, contracts, and transcripts never leave your computer.' The Memory layer that holds your rituals, voice samples, templates, and history is on local disk under ~/.clone/memory/. You can route the Planner LLM to a local Llama 3 on an M-series Mac for fully on-device operation, or to a cloud LLM for faster reasoning. Computer Agent and Memory remain local in either configuration.

Days, not weeks. The reference timeline is in the body of this article: install on day 1, run one observed real ritual on day 2, accept the first proposed pattern on day 4, see the first auto-drafted kickoff on day 5. The clone gets sharper every time you accept or edit a draft, because each accept-edit-reject signal feeds back into the ritual file.

Three. First, the clone is only as honest as your last N runs. If your last 12 kickoffs were rushed, the cloned ritual will be rushed. The fix is to consciously run the next two or three by hand the way you actually want to. Second, edge cases (a brand-new engagement type with no prior pattern) get nothing useful from the clone, you write that one by hand. Third, the cost of letting auto-send run on a ritual you have not reviewed twice is your reputation, so the default review queue is a feature, not a friction. Use it.

The comparison table in this article is the comparison page. Five rows, each row contrasts one operational property of the clone (where the output lands, what it learns from, where the work happens, what 'clone' means in practice, and pricing) against the avatar, voice, and knowledge incumbents. The pricing row sources Solo plan pricing from src/components/comparison.tsx line 179 ($49 per month) and the virtual-assistant baseline from line 188 ($3,000 to $6,000 per month).