Changelog
Seven days of features and improvements from the full Murph archive.
7 updates shipped in this edition.
Improvements
Murph performs fewer internal checks before provider activity, reducing overhead before typing and replies.
Provider requests now use the running container’s local authority instead of an extra remote authorization call. Runtime ownership and usage limits remain enforced.
Background syncing can recover from an interrupted run without waiting for your next conversation.
Routine maintenance can clear an inactive run that was blocking recovery, while active conversations and syncs remain protected.
Imported questionnaire answers, scores and medical history can now retain structured values, statuses and source details.
Unknown dates and missing units stay unknown. Murph can also read RTF text and continue through longer documents, while keeping unclear facts and unreviewed images marked as unresolved. Your original evidence remains in the vault.
Connected-device updates now give slow responses more time to arrive before timing out.
Some requests have a longer wait limit. Updates can still be delayed or fail, and your conversations keep priority.
Your messages can take priority when background syncing is still starting up.
Background syncing can still start on its own. If a session is already running, it can pick up your message directly.
Murph can resume an approved question to another chat after saving progress, reducing requests that stall before their deadline.
Existing sharing permissions, answer destinations, and deadlines still apply.
Murph checks that generated voice memos and songs contain decodable audio before uploading them.
Empty files and incomplete audio headers are caught before sending. Playback still depends on your messaging app receiving the audio correctly.
7 updates shipped in this edition.
Improvements
Corrected images across 123 exercises make poses, movement phases, and side changes easier to follow.
The updates replace mismatched exercise pictures, fill in missing movement phases, and clarify confusing limb positions and labels.
Murph can recover supported measurements, lab results and other structured findings from records previously saved only as source notes.
Your original evidence stays in the vault. Unclear details remain unresolved, and recovered facts track corrections or withdrawals of their source records. Advanced intake imports can also retain assessment categories and links to source records. This does not guarantee a complete medical chart.
Shared contact cards use your active Murph texting number.
Saving from the app waits until your texting line is ready. Cards are not sent when messaging is blocked or the current sending number cannot be determined.
Murph can recover sooner when a connection fails as it starts processing your message.
A failed startup request no longer adds an unnecessary waiting period before processing can resume.
Group updates distinguish stalled wearable delivery from a successful sync check.
When device status is shared, sources with detected delivery gaps can show that they need attention. Missing sleep alone no longer prompts a later report time without evidence that records usually arrive late.
Murph now defaults to GPT-6.1 Sol. Existing OpenAI reminders and automations saved with GPT-6 Sol use GPT-6.1 Sol when they run.
Your schedules and explicit reasoning settings stay the same. You do not need to recreate reminders. Luna, Venice, and custom-provider routing keep their existing behavior.
Murph picks up the language you use in private conversation and remembers it for future voice memos.
Your selected voice stays the same. You can change the default or ask for another language just once; a translation request does not reset your preference.
8 updates shipped in this edition.
Improvements
New voice memos use Eleven v4 while keeping your selected voice.
Your saved voice choice stays in place. Previously queued memos keep their original generation settings.
Murph now retries more temporary connection failures while saving completed background work.
A brief gateway failure gets one automatic retry. If it persists, the existing recovery process still handles the pending work.
When no public goal exactly matches your request, Murph keeps your plan tied to the outcome you asked for.
Related public goals no longer become the source of your custom plan. You still review the plan before it is saved, including whether you want reminders.
Connected-device history can continue in smaller windows when a day contains too many heart-rate samples to collect at once.
Imports keep their existing safety limits and only publish complete windows. Oversized hourly responses remain retryable rather than publishing partial data.
Murph can read your sleep history without first preparing unrelated health summaries.
Sleep answers still use your current wearable records, including corrections. Preparing other health views happens when those views are needed.
Murph skips unnecessary checks and waits when there is no pending background result to handle before your message.
Daily Journal and Personal Patterns checks now use Luna with extra-high reasoning.
These two background checks use the lower-usage model while keeping their existing schedules and instructions.
Settings lists Luna, Sol, and Astra from lowest to highest usage.
13 updates shipped in this edition.
Improvements
Reduced background bookkeeping while Murph prepares a reply.
Independent checks run together, and repeated checkpoint work is skipped while preserving your saved progress.
Murph starts the typing indicator sooner after receiving a text message.
Typing can begin while Murph prepares your reply in private and supported group conversations.
Choose Sol, Luna, or Astra in Settings without sorting through model versions.
Older models no longer appear as choices. Existing saved preferences stay in place until you change them.
Voice no longer appears in the sidebar navigation.
The desktop sidebar and mobile menu now show the same shorter list of destinations.
Medical record imports distinguish portal coverage limits from routine notices and label repeated items more accurately.
More text-based lab results can be saved with their original reference ranges. Dated observations without a supported measurement format are preserved as source notes. Document extraction excludes generic education and provider illustrations from medical facts. Uncertain records remain available for reference.
In private conversations, Murph explains a failed step before safe recovery and can acknowledge a delayed link send.
When the text of a reply is accepted but its separate link send remains unresolved, Murph attempts one brief update while retrying the link. Successful requests stay quiet; updates do not invent an outage or a completion time.
Connected-device updates avoid unnecessary pauses when background work receives a repeated wake-up.
New messages still take priority, while repeated background signals no longer stop a sync before it starts.
Review saved events even when unrelated saved records are malformed.
Event filters and corrections stay intact. Listing events does not require rebuilding unrelated summaries or search data.
Calendar plans stay useful context without automatically prompting an attendance check.
Automatic follow-ups focus on a specific health question. Reminders you explicitly requested are preserved, and older generic check-ins are skipped.
Image replies no longer add a separate image description to the message.
Murph keeps the written reply and image together without automatically inserting the image's alt text as another paragraph.
Murph avoids one cause of unnecessary retry delays when a private iMessage arrives during other account activity.
Account and conversation checks still apply before a message is accepted.
Murph checks booking details before using a lodging destination in your travel context.
When a confirmation does not establish the destination, Murph keeps the confirmed dates and marks the location as unknown. Existing imported plans can be corrected when their booking details support a change.
See your latest import outcome and open lab results from a simpler medical records page.
Import details and connection controls sit together under Manage. Incomplete imports stay visible, and Data & privacy remains within reach.
12 updates shipped in this edition.
Improvements
Murph can finish an active reply while a service update is rolling out.
Routine updates allow the conversation work to finish before the old runtime exits, reducing avoidable restarts and reply delays.
Murph keeps support reports focused on the product problem while leaving out private health and family details.
Garmin history imports combine overlapping requests for steps, distance, active calories, and breathing rate to avoid fetching the same days repeatedly.
Imports keep their full date coverage and can resume after an interruption. Completion time still depends on the amount of history and provider response time.
Murph shares its contact card after your first delivered iMessage, whether you start by texting or sign up through the app or website.
Sharing follows delivered iMessage activity, with at most one automatic attempt per day. Your iPhone controls when it offers to save the contact. Accounts without an iMessage connection wait until one is established.
Murph does less startup work before processing a message when its messaging service starts fresh.
The change keeps the same encryption checks and message handling. Response times still depend on service and network conditions.
Daily Personal Patterns updates can recover from longer temporary service shortages.
Murph can retry the daily review for up to four hours after its scheduled time. Ordinary reminders keep their existing delivery window.
Murph can acknowledge your name and move to your first health question through the quick reply path.
Partial answers and skipped identity details work too. If you already have a health goal or need help with something, Murph continues from that context. Your conversation stays intact, and your details carry forward as you continue.
Murph can save an accepted walking or habit plan around a cue like before lunch, without asking for an exact clock time when reminders are off.
The saved plan stays linked to your goal. Choosing quiet support keeps reminders off, and you can ask to change or pause the plan. If you shift the start, a quiet plan’s relative first-week review moves with it.
Corrections meant for future behavior are saved in the right place before Murph confirms them.
Changes scoped to one task stay with that task; broader reply preferences use your settings. One-off edits leave saved preferences unchanged.
Murph can start answering new messages while shared group data updates in the background.
Sharing continues when a new message arrives, using saved data and your existing sharing permissions.
Murph uses your saved workout duration and repetition instructions more consistently when you report later activity.
A duration you give for the current workout overrides your default. Repetition instructions apply only to the workout and exercise you chose, and a cleared instruction stays cleared.
A voice request can interrupt background work without disconnecting your call.
Murph can answer and continue the conversation on the same call while keeping background work under its existing controls.
12 updates shipped in this edition.
Improvements
If your introduction goes quiet, Murph can check in once to help you pick it back up.
The check runs about fifteen minutes after your introduction starts. It stays quiet if setup is complete, the latest setup question has not been unanswered for ten minutes, or you asked to pause or continue later.
Murph uses GPT-6 Luna for quick opening replies and simple reminders, with GPT-6 Sol for reminders that need more context.
Managed check-ins and insights now use GPT-6 defaults. Existing OpenAI reminders already move to the corresponding GPT-6 model when they run; your conversation model and supported alternative-provider choices stay available.
Murph no longer sets up an immediate follow-up nudge during your introduction.
Share or skip the opening identity question and continue the conversation. Later onboarding recovery keeps its existing schedule.
Reminders and automations saved with GPT-5.6 Luna, Sol, or Terra now use their GPT-6 replacements when they run on OpenAI.
Luna moves to GPT-6 Luna; Sol and Terra move to GPT-6 Sol. Schedules and explicit reasoning settings stay intact, with no need to recreate reminders. Venice and custom-provider selections keep their existing behavior.
Murph can get started sooner when several conversations begin close together.
This reduces avoidable setup delays before a reply. Some conversations may still need extra startup time.
Murph gives your personal conversations priority processing during your first 24 hours, at no extra cost.
Available with supported OpenAI models through Murph. The boost uses your normal AI allowance rate and ends automatically after your first day. Custom providers keep their existing processing.
Messages can start processing sooner after background work has finished.
Murph skips an unnecessary retry delay when handing completed background work back to a conversation.
Messages take a shorter preparation path before Murph starts composing a reply.
Removes repeated setup while preserving conversation context and the order of pending replies.
Murph does less setup when checking for new messages.
Family-sponsored conversations also avoid a repeated allowance lookup. Plans and usage limits keep their existing behavior.
Murph spends less time handing off from one conversation turn to the next.
This removes an avoidable pause when your next message arrives as the previous turn finishes.
When a private conversation leads to a repeated-action plan, Murph offers practical reminders and a check-in to help you follow through.
You can accept, adjust, or decline the support. Murph reuses timing you have already shared and asks about timing when needed before scheduling.
If you delete your account and start again in the same iMessage chat, old setup replies no longer prevent the quick opening questions.
11 updates shipped in this edition.
New features
Murph can cast or remove its own vote in iMessage polls it creates, including a playful pick or a tie-breaker.
Murph uses its judgment about when to participate. It can also notice a majority, everyone voting, or a closed Telegram poll and briefly acknowledge the result when useful. Telegram bots cannot cast native votes, so Murph can share a preference in the chat without changing the count.
Improvements
Group reports keep each person's shared results distinguishable when a profile name is missing or shared by someone else.
Murph can use a safe contact name shared by the group owner or a stable participant label when no name is available. Reports do not expose phone numbers or email addresses.
Workout corrections preserve the original finish time, and workout cards remain readable in older Messages extensions.
Active cards keep their editing controls and completed cards keep their summary. If Murph cannot prepare a verified card, it replies with the saved workout details instead.
Murph takes fewer setup steps when looking up food labels or checking data shared with a group.
Food lookups recover more directly from command mistakes, and shared-data questions can skip a discovery step. Group replies continue to use only data members have chosen to share.
Murph keeps a ready session available while an incoming message is being handed off for processing.
This avoids an unnecessary restart during that handoff. Normal idle cleanup and recovery still apply.
Murph uses GPT-6 Sol by default for managed OpenAI chats, with GPT-6 Luna also available to choose.
Terra selections now use GPT-6 Sol. Other saved model choices stay in place. Choose GPT-6 Sol or the lower-cost GPT-6 Luna in Settings, or ask Murph to switch.
Sign-in emails carry 40% less logo image data at the same resolution.
The light and dark logos stay embedded in the email. Display timing still depends on your mail app.
Large device history imports reuse repeated checks and move through dates with no readings with fewer interruptions.
Every historical day remains covered, with current connection checks before saving data and the same priority for new readings and messages.
Completed actions and reported feelings are saved in the forms Personal Patterns can use, and calendar plans can recover a missing check-in.
Unscored symptoms stay unscored. Existing notification history and mutes are preserved, and recovery does not restart paused or completed check-ins.
Murph reviews existing reminders against current context and repairs supported mistakes.
Morning checks can update stale instructions, follow changed event times, and retire confirmed obsolete or duplicate reminders. Explicit timing choices and delivery destinations are preserved. Outdoor reminders also check current travel context before choosing weather.
New top-ups allocate 80% of the purchase price to AI usage, matching paid plans.
Purchase prices stay the same: $5, $10, $20, and $25 purchases provide $4, $8, $16, and $20 of metered usage capacity. This also applies to Family top-ups, group contributions, and new automatic group refills. Existing balances and purchases already started keep their original usage amounts.