What’s new with Murph

Murph now distinguishes a live volley from catch-up, a joke from a plausible emergency, and a human-to-human beat from an opening that actually wants Murph.

It may wait a few seconds while people are mid-burst, gives human punchlines more stage, and asks one short question when the health context is genuinely unclear. Timing never overrides the room's silence, reaction, or not-for-you rules.

When a call Murph started finishes, Murph now sends the meaningful result back through the chat where it can reach you, in its own voice.

The result is encrypted at rest, Murph treats the call transcript as untrusted, and a replay cannot send the same outcome twice. After Murph connects you to the recipient, it waits for that conversation to end and asks what happened instead of guessing the outcome.

If your subscription lapses, Murph answers on your home line and points you to Subscription controls instead of dropping the text or sending you to a blocked signup wall.

Resuming a paused subscription now carries the saved card onto the invoice, avoids an update the provider rejects while paused, and offers the billing page when the provider is still finishing.

After sign-in, accepting Murph's terms and health-data notices now takes one affirmative action instead of a row of checkboxes, with an equally clear Decline path.

The prompt states that Murph does not sell your health data or train general-purpose AI models on Murph-managed health data. After repeated save failures, it offers a support route instead of trapping you in retries.

When a group runs out of included usage, Murph now explains the pause in its own voice and, when funding is available, gives the whole room a clear way to bring it back.

On iPhone, Add Murph to Contacts now hands off from an in-app browser to Safari instead of letting the embedded window swallow the contact-card download.

The short-lived handoff stays bound to your account and chosen avatar. Setup completes only after Safari actually opens; if the launch is blocked, the picker stays open with a retry.

A leftover folder in experiment storage can no longer disable every managed automation or archive active experiment check-ins. Unrelated reminders keep running while Murph retries a temporary experiment scan failure.

Murph can now rename the chat it is already in even when the group has not started a challenge or created a shared workspace. Updating Murph's own label follows when possible.

Murph now receives a safe description of why voice generation failed, while the maximum script length and generation timeout share one limit so overlong memos do not fail by design.

Provider status and request identifiers reach secret-safe logs, giving Murph and support enough information to distinguish a timeout from a quota or request error.

If a database timeout proves that no query or transaction started, Murph's web service now retries it briefly instead of failing the request.

A failure that may have reached the database is never replayed, so the retry cannot duplicate a completed action.

The iMessage roster now separates people who completed Murph setup from people who never activated, without treating current plan status as identity. A paused or lapsed member is no longer mistaken for a stranger.

If a push-based device source still looks connected but stops delivering data, Murph now records the silence as a source-stalled signal instead of treating every empty sync as a normal empty day.

The check is observation only: it does not disconnect the source, block ingestion, or launch a recovery on its own.

The hero now tells one story from the start: friends enter the chat, the challenge begins, and Murph carries it through standings, a roast voice memo, and the Sunday recap.

The separate solo-demo act and its decorative controls are gone, and the security promises now read in one clean column.

A group's Detail setting is now a hard ceiling on message length rather than a target, covering the whole turn including every follow-on bubble.

A direct question still gets its complete answer, as tight as accuracy allows. What the ceiling cuts is volunteered length: scoring frameworks, multi-topic essays, and background nobody asked for.

Asking Murph to share its card again in a group now posts a new card instead of refusing, with a 90-second guard that only collapses duplicate retries.

Group rooms and unverified outside threads now decline code reviews and work deliverables in one plain sentence, while ordinary conversation and quick general questions stay in scope.

The join screen now preselects exactly the permissions a group or challenge asked for and explains them in one line, so joining is a single confirmation.

Family top-ups now start from each member's manage window, and a finished purchase opens Messages instead of ending at a confirmation with nowhere to go.

A linked Telegram account now completes messaging setup at join instead of asking again how Murph should reach you. Until you send the first message Telegram requires, Home puts Message Murph first and links straight to the chat.

Before the final delete confirmation, you can pick a reason for leaving and add a note, or skip the question in one tap. The answer never gates deletion.

If you answer, Murph keeps the optional feedback only after account deletion finishes and stores it without your member id.

The refreshed privacy, health-data, and health-AI documents ask for one review while your account, connections, sync, and current conversation keep working.

Members whose acceptance was merely out of date no longer hit silent failures: hearting a group join offer, sending a meal photo, and tapping a confirmation inside Messages all keep working while the review reminder is outstanding. The homepage also carries a distinct Consumer Health Data Privacy Notice link.

Murph now sees which reaction you used and reads it as acknowledgement by default, treating it as agreement only when the message it answers asked a single yes-or-no question.

Everyone's messages in a group share one daily bucket, so the cap that was sized for a one-to-one chat doubled to 200. Direct chats keep 100.

Every confirmed participant now gets one short ask for a one-line intro and a photo for challenge comics, and someone who opts in later gets the same ask.

Sharing either one stays optional and the challenge starts without waiting. Setup also asks its next question directly instead of first announcing that setup is about to happen.

A generated image now uploads and sends instead of failing after generation and leaving Murph to report that the upload did not work.

A day with more than one workout now reports the full count and combined duration in replies, group comparisons, and newsletters, with the current day labeled so far.

Account creation, trial activation, and phone-line assignment no longer queue behind one global lock, and a brief database stall shows the ordinary homepage instead of an error.

Accepting your first automatic meal capture now sets up the 9pm local closeout without another setting or prompt.

When direct WHOOP slots are full, connecting now opens the same Apple Health setup guide with its voice walkthrough instead of a separate inline dead end.

A section of the dashboard that cannot load is now left out with one retry notice, instead of taking down the page or recommending you connect devices Murph cannot currently see.

Joining from an emailed invite now locks to the invited address and explains how to switch senders, instead of failing with a mismatch you cannot act on.

Murph now completes both supported requests in the same turn instead of doing one and describing the other as a messaging limitation that did not exist.

The early questions now give concrete examples of a useful answer, repeat your actual goals back to you, and hand off a device connection in Murph's own words.

Murph no longer asks you to come back and say a quoted magic word after connecting a wearable, and it only asks the labs question by voice if you answered by voice.

New repeated-measure runs baseline for 14 days instead of 7, which makes the before-and-after comparison far less noisy. Saved runs keep their original timing.

A newly shared lab report gets one short receipt before Murph preserves and reads it, and a check across several health areas says which areas it is about to read.

Routine single reads stay quiet, and overlapping onboarding work collapses into one update rather than several.

After you approve a secure file, Murph finishes that exact send on its own as one attachment, without a second approval request, a typed confirmation, or a caption bubble.

Asking a group you joined now starts the read-only group turn right away and wakes your own Murph when the answer lands, instead of parking it until your next message.

Hosted containers moved from one to two vCPU with double the memory, which shortens the heavier reads Murph does mid-conversation.

Finishing the first-visit persona and tone step now opens the welcome dialog with a Text Murph action on your resolved channel, instead of ending on a blank page.

The family manager becomes readable cards instead of a clipped table, the customization drawers fill the screen, and every Connect card anchors its action to the same place.

The guide after connecting WHOOP now offers a clear Download App action and a Continue with Murph handoff to your existing Messages or Telegram thread.

At the existing 20 percent threshold Murph finishes what you asked first, then may add one casual sentence that the conversation could pause soon. The separate automated warning is gone.

From an authenticated group chat, Murph collects the name, cadence, delivery target, tone, and health scopes and saves them once, then waits for the natural first occurrence.

Delivery stays pinned to the chat that was authorized. If that thread is replaced or revoked, the occurrence fails instead of sending shared health data somewhere new.

A finished report now draws the saved baseline and intervention measurements day by day, with the summary averages consistent with those points.

Run reports now use the ordinary page margins, one metric chart per row, and the standard trend chart, and finished runs drop the repetitive saved-result narrative.

An authenticated member can now link or replace a phone number without a false signup-completion error, and the dialog opens directly on the focused phone field instead of repeating the Settings status card.

The fix still requires a fresh login for the exact same account and continues to reject a phone identity that resolves to another member.

Murph no longer turns one bad night into an alcohol assumption or sends a weekly insight whose main point is that drinking or a late weekend hurt sleep or recovery.

That conclusion is treated as obvious and unhelpful here, so Murph can choose a different evidence-backed candidate while keeping causal claims appropriately cautious.

An automation now runs as an ordinary conversation turn with your usual prompt, skills, style, and tools, instead of a separate reduced profile with its own thread policy.

Confirming Start Pulse once now completes automatically when you return from Stripe, whether you began in Settings or from a link Murph sent in chat.

Cancelling returns to ordinary Settings without starting billing, and a terminal failure offers one retry you control.

Runs whose source protocol left the public catalog kept their saved data but lost their links. They now open to the progress, measurements, and conclusions already saved in your vault.

A question about a single named marker now uses one targeted lookup that carries the value, unit, flag, and reference range, instead of loading your whole lab history first.

When one memo covers movement, supplements, and medical history, Murph sorts all three at once and says it is working instead of going quiet.

Murph now sees the welcome it already sent when you reply, so it answers your message instead of introducing itself a second time.

The approval screen now offers Sign in to approve as soon as it knows the browser is not authenticated, rather than waiting out a loading timeout.

Murph no longer offers new or renewed Strava connections and says so plainly. Existing connections keep their data, status, and disconnect control.