What's New ​

Here's what's been added recently. If you're coming back after a break, this is the quickest way to catch up.

Fixed: "JSON Parse error" stopping Whisper mid-task ​

On some accounts, Whisper could stop with a red JSON Parse error: Unexpected token ',' message, often right after it started, and then could not finish what you asked (like scheduling a mass message). That's fixed. Run whisper update to get it.

Truer selling numbers, and no reacting to photos that never came ​

  • Offers to one fan count as offers. A paid message you scheduled for a single fan now shows up as an offer in whisper autopilot report, not as a mass message.
  • Older chats read right. In chats saved before late October, some of your own paid messages were filed as if the fan sent them. Autopilot no longer mistakes the media in those older messages for a photo from the fan.

A reject with a reason gets a new draft right away ​

When you reject a card and type why ("too pushy", "he asked about her day"), Autopilot now writes a fresh draft for that fan straight away, with your reason in mind. It does this once per fan message, so a second reject leaves the chat with you. Skip still means "leave this one for a person" and never redrafts.

Learn more about Autopilot

Autopilot fixes from a full night of live chatting ​

  • No fan left waiting behind a mass message. A mass or scheduled message no longer counts as you answering a fan who wrote just before it.
  • One card per fan. When a newer draft for a fan is queued or sent, his older waiting cards close on their own.
  • No empty cards. If drafting fails for a moment, Autopilot tries that fan again a little later instead of showing you a blank card.
  • Fewer false stops on wording. The restricted-word check no longer trips on harmless words that only contain a blocked word, and a draft stopped for wording gets one fresh try.
  • Restarts catch more chats. On startup Autopilot looks back through all chats from the last 3 hours, not just the first 20.
  • New-message alerts reach the right fan even when his chat is already marked read.
  • Your own media first. When a fan sends you a photo or video, the reply reacts to it before any offer, and fans who have never bought get a photo option, not only videos.
  • Nothing promised that is not in your vault. When a fan asks for something specific you do not have, the card comes to you as a possible custom instead of a tease.
  • Safer replies. Drafts never encourage anything sexual in public or around people who have not agreed to it.
  • One paid offer at a time. A paid message waiting for a fan counts as an open offer, so Autopilot does not stack another on top.
  • Nothing from your wall is sold in chat. Photos and videos you have posted on your page are left out of paid offers.
  • Sexting is not a custom order. A fan asking you to moan his name in the moment no longer opens a custom request.
  • Backfill does not hang. A chat that stops responding now times out, and Ctrl+C stops right away.

Fixed: mass messages that looked like the fan wrote them ​

Some mass messages you sent showed up in Whisper as if the fan had written them. That could make a chat look like the fan was waiting on you, so Autopilot sometimes drafted a reply to your own mass message. Whisper now labels these correctly. Chats saved before this fix are being corrected for you; run whisper update, then whisper backfill once, to bring the corrected chats onto your computer.

See how each account is doing at a glance ​

Every card on your accounts page now shows five things at a glance. Each one has a colored dot that means the same thing everywhere: green is working, yellow needs you, and grey is off or quiet. Hover over any of them to see what to do next.

  • Autopilot: whether it is running right now, in which mode, and on which computer. When it is off, you see when it was last on.
  • Approvals: how many cards are waiting for a decision, and how long the oldest has waited. Click it to open the approvals page.
  • Chats: how many conversations Whisper has saved and when they were last updated. It turns yellow after 3 days (run whisper backfill to catch up).
  • Vault: how many of your photos and videos Whisper has described, with a progress bar. It turns green once nearly all of them are done (run whisper vault caption).
  • Instant alerts: when the last new-message alert came in, if your account has them turned on.

Explicit replies can earn the right to send on their own ​

Approving every explicit reply is a lot of keypresses. Start the cockpit with --earned-autonomy (in bounded_auto mode with --auto-send) and Autopilot sends plain explicit replies by itself, but only once your own decisions have earned it: at least 30 explicit cards decided in the last 14 days, 9 out of 10 of your latest sent as written, and no recent "does not sound like me" or "too pushy" rejects. About 1 in 10 still comes to you as a card. Money, customs, writing first, your biggest spenders, and risky fans always stay with you. The cockpit shows every message it sent on its own, and if one should not have gone out, whisper autopilot flag sends everything back to cards until it is earned again.

Learn more about Autopilot

Autopilot can see your saved media ​

Until now, Autopilot only knew your photos and videos by their folder names. Run whisper vault caption once and Whisper looks at each item and writes a short private description of it. From then on, offers match what each fan has said he is into, the message that goes with an offer fits what is actually attached, the free preview is one that tempts without giving everything away, and offer cards tell you in a few words what is attached. Your folders still win whenever they say something. After the first run, new uploads are described on their own while Autopilot is running.

Learn more about Autopilot

Autopilot learns to sell ​

Autopilot now does the part of the job that makes money, the way a great chat partner would, and you stay in charge of every decision that matters.

  • Offers that feel human. Autopilot offers paid content when a fan asks for it or when the chat has been heating up, not on a timer. It picks content from how you already organize your saved media, never offers something a fan already owns, prices from what that fan has paid before, steps up gently during a hot session, and never chases a "no" with a discount. For fans who are actively unlocking, it can suggest a locked message where the text itself is the paid part.
  • Better approval cards. Offer cards show the price, what is attached, and why. Change the price with P, swap the content with M, and reject with a one-key reason.
  • Session budgets. Press B on an offer to send it and let Autopilot keep going with that one fan for 3 hours (up to 4 more offers, under a cap you set). The first offer always asks you.
  • Approve from your phone. The new approvals page on your dashboard shows every waiting card. Send, edit, change the price, start a budget, or reject from anywhere. Autopilot on your computer carries it out with the same safety checks, and hands the card back if the fan wrote again in the meantime.
  • Your team's notes come with you. Labels you or a former chat team left on fans (nicknames, "don't pitch PPV", "do not message", birthdays, chargeback warnings) are now read and followed.
  • A customs desk. When a fan asks for a custom, the card says so. Press C to quote it and open an order. Whisper spots the payment, reminds you as the due date gets close, and drafts the hand-over when you run whisper customs deliver. Set a rate card once with whisper customs rates and quotes come pre-filled. There is a customs board on the dashboard too.
  • Writing first. A few times a day, Autopilot suggests reaching out on its own: a buyer who went quiet, a birthday, a check-in after an unlock, a custom update. Each one is an approval card, never sent automatically, at most 8 a day.
  • A selling report. Run whisper autopilot report to see unlock rates by price, how your offers to past buyers did, how your cards were handled, and what Autopilot earned, next to the same numbers across all your accounts.

Learn more about Autopilot

Steadier syncing, and active fans first ​

A round of reliability fixes. When things get busy for a moment, Whisper now waits and tries again instead of stopping, so history pulls finish and your saved media library no longer comes up short after a hiccup. Long lists of fans, chats, and media are now read all the way through.

When you ask Whisper to list your subscribers, it now shows your active fans by default. Before, the list could fill up with fans whose subscriptions had ended. You can still ask for expired fans, or everyone, any time ("show me my expired fans"). Searching for a fan by name still looks across everyone.

Tell Whisper why you rejected a draft ​

When you reject a draft in the Autopilot cockpit (or in whisper autopilot review), Whisper now asks one short optional question: why? Press Enter to skip it, nothing slows down. But if you type a few words ("too pushy", "wrong tone for her", "she already bought this"), that reason becomes a lesson, exactly like your edits already are. Whisper uses it when it refines your voice, so the same kind of miss stops showing up. Silent rejects still work; rejects with a reason just teach a lot more.

Learn more about Autopilot

The same safety net on every send ​

Campaign mode and the /unreads review flow now run every outgoing message through the same final safety pass as the rest of Whisper: drafts are tidied up before you see them (so what you approve is exactly what goes out), restricted words are caught, and rapid repeat sends to the same fan are spaced out. You will occasionally see a small note when something was cleaned. Nothing to configure, and nothing changes about how you approve, edit, or skip.

Backfill no longer looks frozen on brand-new accounts ​

On a brand-new account, starting a history pull could sit on a still screen for a bit while Whisper finished getting set up in the background. It was still working, but with nothing on screen it looked stuck. Now you see a clear "getting ready" line with a running clock the moment you start, and the pull kicks off as soon as the essentials are ready instead of waiting on everything. If you ran into this on a fresh account, it is fixed, and there is nothing to change on your end.

Autopilot tunes itself to sound more like you ​

Autopilot now improves the way it writes for you on its own, with nothing to set up. Every edit you make to a draft is a lesson. On a regular schedule, Whisper tries out a more refined version of your voice, checks it against your recent edits, and only keeps it if it genuinely would have matched what you actually sent more closely. If a change would not help, Whisper keeps the voice it had, so it never gets worse. This same bar now guards every refresh of your voice, not just the scheduled tweaks: any time Whisper builds a fresh read of how you talk, it has to prove it would have matched your recent edits more closely before it takes over, and a newer account without much history yet gets the fresh version right away. Over time the drafts you approve need fewer edits. There is nothing to turn on, no notification, and no settings to manage. The best way to help it along is to edit drafts that are close instead of rejecting them.

Learn more about Autopilot

Removed: the old auto mode ​

The old whisper --auto command (a one-shot unattended pass over your unreads) has been retired. It was an earlier take on unattended drafting and predates the safety machinery we built for Autopilot. If you used it, switch to whisper autopilot start: same idea (Whisper watches the inbox and drafts replies for you), with explicit approval for every send, safe multi-operator coordination, and clear handling of ambiguous send failures. If you still type whisper --auto out of habit, Whisper now prints a one-line pointer to Autopilot instead of running.

A sound when an approval needs you ​

The Autopilot live cockpit now plays a short attention sound whenever a new draft surfaces for your approval, so you can step away and come back when Whisper actually needs you. Suppress it with --no-bell on whisper autopilot start, or by setting WHISPER_AUTOPILOT_NO_BELL=1 in your environment. Silent mode and no-TTY runs (cron, SSH without a terminal) skip the sound automatically.

Autopilot: a live cockpit for inbox triage ​

A new mode of working with Whisper. whisper autopilot start opens a live cockpit that watches your inbox, drafts replies, and lets you approve, edit, reject, or skip each one with a single keypress. Built for safe unattended drafting and for teams: multiple operators can run the cockpit on the same account at the same time without sending the same message twice.

Highlights:

  • Three modes. shadow evaluates without queueing, draft_assist queues every draft for your approval (the default), bounded_auto lets routine non-explicit replies send themselves while still routing anything sensitive through you.
  • Silent mode. Press Ctrl+D to keep Whisper drafting in the background without inline prompts. Drain the backlog later with whisper autopilot review.
  • Adult, explicit, PPV, custom, and priced content always require approval. Even in bounded_auto. No override flag. See Autopilot for the full reasoning.
  • Safe multi-operator coordination. Locks plus per-approval claims mean two operators pressing Approve at the same instant on the same draft results in exactly one message reaching the platform.
  • Ambiguous send errors never auto-retry. If a send hits a network error after the platform may have already accepted it, the cockpit flags it [FAILED — verify in OF before retrying] and tells you to verify in the chat directly before doing anything else.

Learn more about Autopilot

Account name in your Terminal window ​

If you run Whisper on more than one account, each window now shows the account name in its title and tab. Open one window for one account and another window for another, and your Terminal's tab bar, window list, and dock previews will tell you which window is which. No more squinting to figure out where you're typing.

Backfill: chats, profiles, and clearer progress ​

/backfill, whisper backfill, and the first-time history offer now finish with a chat and profile pass for subscribers in your lookback window (recent inbox activity plus anyone who spent in that window), so your local cache and cloud backup line up with what deep analysis expects. The Terminal shows three stages (earnings, vault, then chats), with a running clock, which fan is in progress, and message counts so long runs never look frozen. After a few fans finish, Whisper also shows a rolling ETA for the chats stage that updates as it learns your account's pace. After sixty minutes the timer switches to hours, minutes, and seconds. Press Ctrl+C once to stop gently; run backfill again later to continue.

Deeper data for cloud analysis ​

Deep analysis can now pull more from your synced cloud cache when it needs it: top-performing posts, mass-message performance, per-fan profile snapshots, and per-fan chat exports when those files exist for an account. Low-activity conversations might not have a cached export yet, so the tool can still fall back to live chat history when needed.

First-run history pull and /backfill ​

When you connect an account that does not have any transaction rows cached yet, Whisper may offer to pull about the last 90 days of earnings, vault catalog, then chat history and profiles for people in that window, with a progress view for each stage. You can skip and pull later with /backfill or whisper backfill (optional --days on the Terminal command). Deep analysis (/analyze) also gets a short note about how far back your cached transaction data goes so answers stay honest when someone asks for a longer window than you have synced.

/analyze can hand back files ​

When you ask /analyze for a CSV export, a written report, or any other file, it now downloads the result straight to your computer. By default everything lands in ~/Downloads/whisper-analyze/ (a new dated folder per run), and Whisper prints the full path so you can find it. Use /downloads to see the current location, /downloads <path> to point it somewhere else, or /downloads reset to go back to the default. If /analyze doesn't produce any files, nothing changes.

Better deep analysis output and more accurate cross-device data ​

The /analyze command in interactive mode now shows a clearer running view: you can see each step as it happens (which tool is in flight, what command is running, whether it succeeded or failed), and the final answer renders with proper formatting (bold, lists, tables) instead of plain text.

Behind the scenes, deep analysis is also more accurate. If you have multiple operators on the same account, or you switch between devices, your local copy now picks up the latest activity that another device captured, even when the data hasn't been compacted yet. And earnings sync no longer skips slow-moving categories (like tips) when faster categories (like chat purchases) advance the cursor, so per-category history stays complete.

Net effect: rankings (top tippers, top spenders, top buyers) are correct on the first try, and the answer renders cleanly.

Grounded numbers and cloud analysis ​

When you ask about subscriber counts, revenue, or other stats in the chat, Whisper can double-check figures against live account data before you see them. Verified numbers get a quick check mark. If something cannot be confirmed, it is tagged so you are not misled. You can turn verification off or switch stricter handling with /grounding at the prompt.

For bigger analytical questions (patterns, comparisons, anything that needs digging across several data sources), use /analyze with your question in interactive mode. Whisper runs a deeper pass in the cloud and saves a transcript under your account folder so you can read it again later.

Startup flags: --strict-grounding keeps uncertain numbers from showing as if they were facts. --disable-verifier or --no-verifier skips numeric verification for that session if you need to move fast.

Unlimited Earnings History ​

Whisper no longer caps how many transactions it fetches per type. Previously, each transaction type (tips, messages, subscriptions, purchases) was limited to 500 records, which meant high-volume accounts could lose weeks of historical data. Now Whisper fetches everything within the lookback window (default 90 days, configurable up to a year or more).

You can also expand the lookback window for new account onboards to get a complete historical picture from day one. More data means better scoring, better strategies, and more accurate recommendations. If you work on a team, only one operator needs to trigger it. The data syncs automatically.

Learn more about the earnings history window

Purchase History Tracking ​

When you look up a subscriber, Whisper now shows which PPVs they've actually opened and bought, including media types, dates, and content details. The AI uses this to avoid recommending content a subscriber has already purchased and to suggest follow-up content based on what they actually buy.

This is especially useful for high-value subscribers with long purchase histories. Ask things like "what has this person bought before?" or "pull up their profile and purchases" and you'll get a full breakdown.

Cloud Sync ​

Your data now syncs to the cloud automatically in the background. Chat histories, earnings, vault catalog, subscriber profiles, conversation summaries, style learning data, directives, and campaign logs are all backed up as you work.

Switch devices and pick up exactly where you left off. If your laptop dies, nothing is lost. Start a new session on any machine and everything is right there. For teams, multiple operators on the same account see each other's work without any setup.

Startup stays fast because chat histories load on-demand, only when you open a specific subscriber's conversation. Everything else syncs quietly in the background.

Mass Messaging & PPV Campaigns ​

You can now create, schedule, and manage mass messages and mass PPV campaigns entirely through natural language. Tell Whisper what you want to send, who to target, when to schedule it, and at what price. Whisper handles the rest.

  • Schedule messages for specific dates and times or send immediately
  • Target subscriber lists, specific users, or exclude lists
  • Set PPV pricing ($3-$200) with media from your vault
  • View, edit, or cancel scheduled campaigns
  • Dry run mode to preview before committing
  • Campaign performance tracking with buyer drill-down

Works with the subscriber list management below to build full targeted campaigns.

Subscriber List Management ​

Create, rename, delete subscriber lists, and add or remove members, all through conversation. Build targeted segments for your mass messages without leaving Whisper.

Use natural language to find subscribers by interest, topic, location, or any detail from past conversations. Whisper searches through saved subscriber profiles (preferences, notes) and conversation summaries to surface matches.

Examples: "find subscribers who are into cosplay", "who mentioned their birthday", "find fans in California".

Revenue Autopilot ​

Every session now starts with a quick revenue briefing: earnings trend vs. last week, recent campaign results, and pending scheduled messages. Whisper uses this context to proactively suggest campaign ideas, timing, and pricing based on what's worked before.

Time Awareness ​

Whisper now knows the current date, time, and your timezone. This means better scheduling, time-sensitive responses, and accurate "when was the last time" answers.

Query earnings with period-over-period comparison. Ask "how are earnings this week vs last week?" and get a side-by-side breakdown with percentage changes. Revenue chart data from the platform is also available for visual trends.

Enhanced Subscriber Profiles ​

The subscriber detail view now includes recent transaction history and a computed segment classification (whale, high_value, dormant_whale, etc.) alongside the existing profile data and conversation summary.

Vault Intelligence ​

Browse vault media with tag filtering and performance cross-referencing. See which media has been used in successful campaigns so you can pick proven sellers.

Windows Support ​

Whisper now runs on Windows in addition to macOS. Install from PowerShell, and everything works the same: login, chat mode, updates, all of it. See Installation for the one-line install command.

Dashboard ​

Whisper now has a web dashboard where you manage everything: connect accounts, invite team members, monitor active CLI sessions, and revoke access when needed. This is also where new users sign up and get approved.

Learn more about the dashboard

Browser Login ​

No more API keys or manual configuration. Run whisper login and it opens your browser. Sign in, and you're connected. Accounts you've linked in the dashboard are automatically available in the CLI.

Learn more about setup

Team Collaboration ​

Multiple operators can now work the same account at the same time. Whisper automatically distributes conversations so nobody steps on each other. If someone is already talking to a subscriber, that conversation is filtered out of your unread list. If you do message someone another operator is active with, you'll see a heads-up.

Learn more about working as a team

Auto-Seeding ​

Scoring data is now built automatically on your first startup. No need to run whisper insights before you start getting conversion scores. Whisper seeds the data in the background while you work.

Continuous Learning ​

Every time a subscriber makes a purchase, Whisper automatically learns from the conversation that led to it. The scoring system improves with every sale, no manual runs needed. Think of it as the system getting smarter in real time.

For the richest learning, run whisper insights --deep at least once. This gives the system far more conversation history to learn from, which means better scoring and better recommendations going forward. If you work on a team, only one person needs to run it. The scoring data and conversation history are shared across all operators on the same account automatically.

Learn more about insights and scoring

Multiple Accounts ​

New guide covering how to add newly linked accounts, switch between accounts, and run several accounts at once. If you manage more than one account (or just added a new one), this is the page to read.

Learn more about multiple accounts

Three-Memory Architecture ​

Whisper now runs three distinct memory systems: per-subscriber intelligence, session memory, and a permanent skill layer that learns from your corrections and never forgets. This is the foundation that makes every other feature work better over time.

Learn more about the three-memory architecture

Standing Directives ​

Set instructions that apply to every conversation. Things like "Push the Beach Day set" or "Valentine's promo ends Feb 28." Whisper follows these automatically, and it also creates its own directives when it spots patterns across conversations.

Learn more about directives

Subscriber Routing ​

Jump to any subscriber's context with /sub alice or send a one-off message with @bob check his latest message. Tab completion makes it fast. No more hunting through unreads to find someone specific.

Learn more about subscriber routing

Per-Subscriber Memory ​

Whisper now remembers each subscriber individually across sessions. Conversations with Alice don't bleed into Bob's context. The AI maintains separate memory, profiles, and history for every subscriber.

Style Learning ​

Edit drafts in /review mode and Whisper learns your writing voice over time. After about 5 edits, it starts auto-generating style rules. After 20, it writes like you.

Learn more about managing unreads and style learning

Conversion Scoring ​

Unreads and review mode now rank conversations by conversion opportunity. Higher score means more likely to buy. You always work on the most valuable chats first.

Contextual Tips ​

The prompt now shows helpful tips before each turn, adapted to where you are in your workflow. New user? You'll see basics. Experienced? You'll see power-user shortcuts.

Smarter Session Recovery ​

If a conversation gets really long and hits a context limit, Whisper automatically compacts it and keeps going. You'll see a brief "session compacted" message, but nothing is lost. No need to restart.