AI meeting follow-up emails that wait, reconcile, then draft.

Three hours is a default because most consulting commitments that can ship same-day ship in that window. A proposal revision takes an hour or two. A deck edit takes 30 to 90 minutes. An intro forward takes a minute. A calendar invite for the next meeting takes seconds. Three hours after a 10:57 call means the reconcile fires at 13:57, which is long enough for the realistic same-day work but short enough that the draft still lands before the client's own end of day. You change it in line one of memory/rituals/meeting-followup.md. For high-velocity sales calls where the client expects a recap faster, 30 to 60 minutes works. For strategy calls where the deliverable is a week out, you can move it to 24 hours and Clone will still correctly branch to the none_promised or partial_ship case.

It watches for new files saved to your Drive (or desktop, or a specific client folder) in the window from 0 to the end of post_meeting_window. It matches on three things: (1) the file name against artifact names mentioned in the transcript, (2) the file path against your client-folder convention, and (3) the file type against the artifact type the transcript promised (deck, SOW, report, doc). A match on two of three is enough to flip the draft into the artifact_ready branch. The exact file is attached to the draft and a one-line note in the reconcile report says where it came from. If nothing matches, the check returns empty and the next branch logic takes over. This is the same Computer Agent observation principle described in src/components/architecture.tsx lines 18 to 22: Clone reads the screen, clicks, types, and scrolls the same way you do, so a file you saved is visible to it without any integration glue.

Fireflies and Otter both compose an email from the call transcript and fire it within about two minutes of the call ending. The email is grounded in what was said on the call, nothing else. Clone waits the post_meeting_window (default 3 hours), then reconciles the transcript's promises against five surfaces (Drive, Gmail, calendar, Slack, CRM), then picks one of four draft branches based on what the reconcile found. So the same call with the same transcript produces a completely different email depending on whether you shipped the artifact, already forwarded the intro, or let the deliverable slip past your promise. Fireflies and Otter are good at transcription. They don't reconcile anything, because nothing has had time to happen yet when they fire.

It uses whatever transcript source you tell it to. Zoom's native transcript, Google Meet notes, Microsoft Teams transcript, or a third-party tool (Fireflies, Otter, Fathom, Grain, Read.ai, tl;dv, Avoma). Change the source_of_transcripts line in meeting-followup.md if you switch. Clone does not re-transcribe anything it doesn't have to, and it does not charge you for transcription minutes. If you want to drop your transcript tool entirely and use only Zoom's native captions, that works too. The reconcile logic is independent of the transcript source.

Each call gets its own reconcile timer. A 10am call fires at 1pm. An 11am call fires at 2pm. A noon call fires at 3pm. Six calls spread across the afternoon produce six drafts spread across the late afternoon and evening, each one composed at the point where enough post-call work has happened to reconcile cleanly. The drafts do not pile into the same 10-minute window like they do with transcript tools that fire immediately. You also see each reconcile report separately, so if one call had an overdue artifact, you notice it in isolation instead of hidden inside a batch.

Yes. The reconcile checklist is a declarative list in meeting-followup.md. Delete or comment out the surfaces you don't use. Most solo consultants run four of the five defaults (Drive, Gmail, calendar, CRM), drop Slack, and have the exact workflow they want on day one. Adding a surface later means adding a new stanza with a match_on block. The draft_branching logic runs against whatever surfaces returned a status, so a three-surface reconcile still produces four valid branches, it just has less evidence to attach.

It drafts a short, honest email that acknowledges the missed promise and sets a realistic new time. The voice source is your last six 'running late' sent emails, so the phrasing mirrors how you actually apologize for slipping a commitment (shorter, less apologetic than you think, always with a specific new date). It stays as review_first by default, which means it queues in Gmail drafts and never sends without you reading it. In practice, this branch is the single highest-value output of the whole system for most consultants, because it catches the deliverable slip before the client notices and prompts you to either ship it fast or write the honest update.

No. The reconcile report is a block of markdown Clone pins below your draft in the Gmail compose view, for your eyes only. It lists every check (drive, gmail, calendar, slack, crm), the status of each, and the evidence (the file path, the forwarded thread ID, the calendar event ID, the stage change timestamp). When you hit send, the report is stripped out and only your email body goes to the client. You can delete the report before sending if you don't want to see it, but most users keep it because it's the record of why the email has the shape it has, and it makes post-call audits trivial.

The consulting-follow-up-automation guide is about the content side: lifting the specific commitment from the call transcript and composing it in your voice, one message. This guide is about the reconciliation layer on top: waiting a window, checking what actually happened on five surfaces after the call, and picking which of four drafts to write based on what you shipped. The two rituals (followup.md and meeting-followup.md) coexist in memory/rituals/. The content ritual says how a single follow-up is phrased. The reconciliation ritual says when it fires and which branch it is. Most practices run both files together.

Both. Sales calls tend to want a shorter post-meeting window (30 to 60 minutes), because the artifact promised on a sales call is usually a deck or a one-pager that goes out same-hour, and the client expects it fast. The branches still apply: artifact_ready when you attached the deck, partial_ship when you attached the deck but the pricing page is still being finalized, artifact_missing_overdue when you said 'I'll have this in an hour' and the hour passed, none_promised when the call was exploratory. Many sales teams run meeting-followup.md with post_meeting_window set to 45 minutes, and the artifact_ready branch fires on most calls where the rep shipped the follow-up deck right after the call.

No. Every draft Clone produces is a native Gmail draft in your own mailbox. When you hit send, the email flows out of Gmail's native send, subject to your Gmail sending limits, your SPF / DKIM / DMARC, your IP, your sender reputation. There is no third-party sending infrastructure involved. This is strictly better for deliverability than transcript tools that pipe mail through their own servers or inject tracking pixels. Gmail's default 500-per-day limit for Workspace is an order of magnitude more than the 3 to 8 post-meeting drafts you'll actually ship in a day.

$49 per month on Solo, one flat plan, 21-day free trial. Compare against Fireflies Business $19/seat plus the AskFred and CRM add-ons, Otter Business $20/seat (per minute caps apply), Fathom Team $29/seat, Grain Business $29/seat plus the sales tier, Read.ai Pro $19/seat. Most of those require a separate CRM add-on seat to make the follow-up loop complete, and most of them charge per active contact or per transcription minute above a threshold. Clone is one flat plan with no per-meeting or per-contact fee, and the drafts live in Gmail so you are not paying for a second sending infrastructure.