Changelog
Seven days of features and improvements from the full Murph archive.
28 updates shipped in this edition.
Improvements
Murph can check in privately on iMessage after three days without an Apple Health update, with a reminder to open Murph and check for new data.
You'll get one check-in per gap, in an existing active conversation. Ask Murph to wait 5–30 days or stop these check-ins; your connection and syncing stay on. A gap alone does not mean your app was closed or your connection broke.
Edge and Max members can choose GPT-6 Astra for their conversations with Murph.
Available on individual and Family Edge or Max seats through OpenAI. Choose Astra in Settings or ask Murph to use it for future queries. Astra uses your AI allowance faster than the other models.
Murph recognizes supported services that do not need you to connect an account.
If a built-in service is temporarily unavailable, Murph distinguishes that from a personal account that needs connecting.
Ask Murph to follow activities from Garmin, Oura, or Fitbit when setting up an activity check-in, or change the device on an existing check-in.
Check-ins wait for matching activity to sync from the selected device. WHOOP filters remain available, and check-ins without a selected device continue to follow all activity sources.
Connected-health refreshes keep their saved progress as new sync details arrive.
This fixes a case where a refresh could repeatedly retry instead of finishing.
Murph makes better use of ready capacity to reduce avoidable waits when you return to a conversation.
More than one conversation can start from ready capacity at the same time. Each restores its own saved context first; ordinary startup remains available when ready capacity is busy.
Murph can resend a reply once when iMessage confirms that the original send failed.
Recovery keeps the same conversation and content, respects current messaging restrictions, and stops after one attempt. It covers text, links, and non-audio attachments; voice memos, interactive cards, and uncertain delivery failures keep their existing behavior.
Murph keeps the opening questions moving while saving the name and personal details you share in the background.
You can skip a question or ask for help at any point.
Murph can send requested PDFs and other files as attachments in your private Telegram conversation.
Review and approve the file and destination before Murph sends it. If delivery cannot be confirmed, Murph avoids automatically sending a duplicate.
Murph can answer individual workout questions and verify saved preference changes without loading unrelated detail.
Workout summaries retain timing, distance, provider, and other workout facts. Lap and split details remain available when your question needs them, and saved preference changes still use canonical records.
When you ask Murph to help with a concrete health outcome, it now checks for matching public Goal guidance before beginning the setup conversation.
Informational questions remain ordinary health Q&A, and Murph does not save a Goal or ongoing support until you accept the proposed plan.
When you ask Murph to connect a reminder to a saved condition, it can now load the record-reference format before creating the reminder.
Murph uses the canonical condition identifier and creates the reminder once, avoiding a failed save caused by guessing the reference fields.
When you first message Murph through an iMessage email address, an approved conversation can begin with an answer instead of a signup link.
The address keeps the iMessage conversation attached to the right account. It is not treated as a verified account email until you verify it.
When a saved wiki page is missing, Murph can explain that it is unavailable and move on without repeating the same failed lookup.
Asking to read a missing page does not authorize creating a replacement.
When Murph summarizes an idle conversation, usage accounting now uses the provider's reported input, output, and cached tokens when available.
Conversations without exact usage reports retain the existing estimate. Recorded usage may be higher or lower than that estimate.
When you ask for a daily nutrition card, Murph checks your saved meals and applicable nutrition targets together.
Conflicting or incompatible targets still produce an explanation instead of a card, and the check does not change your goals.
Murph skips empty background checks so more of your usage stays available for useful help.
Memory checks can stay idle when there is no conversation to review, and Journal checks can stay idle when no accounts are connected and no earlier Journal work needs follow-up. Existing context, new connections, and scheduled support keep their normal checks.
Murph retries a stalled context download when reopening your conversation, reducing waits caused by an inactive download.
Downloads that keep making progress can continue, and an unsuccessful recovery preserves the saved workspace.
Ask Murph whether a reminder ran, is retrying, or is waiting for delivery.
Murph can explain recent attempt outcomes and recorded failures. A queued or failed attempt is never treated as confirmation that a message was sent.
Murph can keep a reply attached to an earlier question when another message arrives while it is answering.
This works in supported private and group iMessage and Telegram conversations for messages received during the same active exchange.
WHOOP members can authorize Murph directly from Connect again under the provider's expanded allowance.
The existing post-connect guide still points WHOOP members to the Murph iPhone app and Apple Health for step data.
Telegram sign-in and account linking now use Telegram's current secure authorization flow.
Returning from Telegram now finishes through the Telegram path instead of showing unrelated phone recovery. Phone and email sign-in are unchanged.
Murph handles requests to package your vault files as a ZIP in your private iMessage conversation more directly.
Creating the archive leaves the original files in place. The existing approval step still comes before the attachment is sent.
Murph is less likely to report that video analysis failed after it successfully found an answer.
This applies to video questions in private and group conversations. If analysis returns no usable observation, Murph still says so.
Ask another question about a video from earlier in the same conversation without sending it again.
Videos can now be used for follow-up analysis for 72 hours, while images and audio keep their existing 14-day media window. Follow-up analysis works in private chats and supported iMessage and Telegram groups. Clips already removed under the earlier policy need to be sent again.
Murph can offer connection help in your private text conversation after five days without data from an established WHOOP connection.
The check-in applies to existing direct WHOOP connections, including confirmed authorization failures, when you have been chatting recently. Murph asks once per data gap. You can ask it to wait 5–30 days or stop these check-ins while keeping your connection and syncing on.
Murph can recover a clearly defined workout from a saved reminder when you reply with a completed set.
Later set updates stay with that workout. Unreported sets stay empty, and Murph still asks when the saved routine is ambiguous.
Workout setup now rejects misplaced set counts and other exercise details before saving a plan.
This catches cases where a set count became part of an exercise name and left the plan at one set. Valid exercise names, notes, and optional set defaults keep their existing behavior.
8 updates shipped in this edition.
Improvements
Long-running connected-health syncs now keep their recovery path after older private queue content reaches its retention limit.
Retired content stays deleted. Murph continues only work already accepted for processing and does not recreate old payloads or duplicate completed work.
When an assistant standby has gone idle, Murph now replaces its stopped assignment instead of treating it as still warm.
A standby that is still warm remains reusable for the same member across conversations and groups. Murph replaces it only after proving that it stopped; uncertain or mismatched ownership continues to retry safely.
New members who start with email now receive Murph's first welcome instead of waiting while the same delivery retries.
Delivery still uses the member's current verified email and keeps the existing first-contact safeguards. If the member has already started a direct email conversation while delivery recovers, Murph does not interrupt it with a late first-time welcome.
Repeated wearable names now group into short, consistent labels across Journal and Personal Patterns.
Journal hides missing activity values and repeated details, and keeps its private-chat helper aligned. Patterns keeps labels compact and its sortable headers free of extra help controls. Weekly insights can also use verified timing and combination evidence from Journal records.
Murph no longer sends recurring biweekly product notes. You can still ask what is new whenever you choose.
Direct questions continue to use Murph's current changelog and feature catalog. Other reminders and health check-ins are unchanged.
When you ask for a reminder in the morning, afternoon, or another broad window, Murph can use your known routine and connected calendar to choose a specific time that fits.
Exact times stay exact. Calendar checks use an existing connection, and the saved reminder remains a fixed time you can change.
When you ask Murph about a video, the analysis now uses Google's latest Flash model while keeping standard and detailed-motion sampling.
Each request remains one bounded analysis call. Older in-flight requests stay compatible during the rollout.
Murph now keeps the relevant workout context when another topic appears before you report a completed set.
A short completion reply can resolve and log the intended set without making you repeat which routine you are doing.
7 updates shipped in this edition.
New features
The Food page compares up to four exact products and marks the lowest or highest value for calories, protein, sugar, and fat.
Linked lab results and missing evidence stay separate from nutrition. Missing evidence is not a failed test or a safety verdict.
Settings now lets you replace Telegram or remove a linked Telegram, email, or phone sign-in when another supported sign-in remains.
Murph confirms the provider disconnect before updating messaging access, preserves unrelated channels and billing records, and keeps cleanup available after a refresh if it is interrupted.
Improvements
Queued connected-health updates now resume when an account-channel refresh is waiting ahead of them.
Both updates keep their durable ordering and retry behavior, while recovery no longer depends on sending Murph another message.
Murph now favors next-day timing for supported wearable summaries and gives a clearer recovery step when Apple Health data is late.
Exact requested times and action-timed reminders stay unchanged. In a group, Murph names a participant only after consented shared status shows Apple Health is their sole connected source; missing data remains unknown rather than zero.
Connected health data can finish syncing sooner after a background refresh completes.
Murph removes an unnecessary final wait once a refresh has safely finished, while preserving normal connection cadence and recovery when work is still pending.
Routine connected-health status checks now release unused backend resources sooner, reducing avoidable delays.
Active syncs and authenticated requests continue unchanged. The improvement only shortens cleanup after database activity becomes idle.
Established group conversations now skip avoidable setup work before Murph starts a reply.
Previously resolved speaker names can be reused after a restart, while first-time group setup still refreshes the information needed to activate the conversation.
5 updates shipped in this edition.
Improvements
When a message arrives as background work finishes, Murph now starts the reply as soon as that work releases instead of waiting for an older retry timer.
The handoff is matched to the exact finished runtime, so a delayed callback cannot interrupt newer work; durable reconciliation remains the recovery path if the handoff signal fails.
Murph now recovers when connected-device work overlaps an outdated assistant timer, the behind-the-scenes cleanup after a reply, or newly reported health data.
Current conversations, reminders, and pending deliveries keep priority. Completed replies finish cleanup without sending duplicate text, while queued connected-health updates continue in order without needing an unrelated message to restart them.
Murph now resumes delayed connected-health and routine background work in order instead of repeatedly waking without progress or letting an older retry delay fresh device data.
Current conversations and approved actions still run first. Existing provider retries keep their recovery timing, while a fresh device update can start a bounded sync pass without needing an unrelated message.
Journal keeps seven-day context close to the timeline and hides empty or repeated activity details. Patterns uses shorter labels for recognized factors.
On phones, the seven-day summary appears before the timeline, calendars and metric trends open from the bottom, and pattern insights follow the period they describe. Useful workout, travel, laboratory, and experiment details stay one tap away.
When several quick messages arrive while Murph is composing in a group conversation, Murph now considers the complete beat before sending one reply.
Only consecutive messages in the same group can join the active response. Unrelated conversations and messages that arrive after the reply window remain separate.
7 updates shipped in this edition.
New features
Journal now shows a rolling seven-day timeline and can add clear life context from new calendar, travel, and Environment changes.
New calendar and email connections get a private notice before use. Medical and private events stay excluded, travel stores only normalized itinerary facts, and every automatic entry remains correctable through Murph.
Murph can privately ask before saving a clear fact that comes up in an authenticated group.
The first clear fact asks for one global choice in private. Later clear facts can save quietly, unclear facts get one private question, and each group can be excluded. Nothing about another member's Journal appears in the group.
Improvements
Murph now keeps simple fixed cues on the lightweight path and uses the more capable model when a reminder needs current context, interpretation, or judgment.
Straightforward fixed cues stay quick. Reminders that need history, current data, tools, personalization, ambiguity resolution, or safety judgment use Terra; existing saved reminders keep their current model unless their instructions or context requirements materially change.
Automatic meal cleanup now stays silent when it is processing captures from an earlier day, so an old nutrition recap does not appear as today's closeout.
Murph still reviews the saved capture and removes the retained photo through the existing privacy cleanup. Current-day closeouts keep their normal behavior.
When an internal command has an invalid or missing value, Murph now gets clear correction guidance instead of a generic failure.
Malformed requests stop before the command runs, and recovery messages do not repeat submitted values.
If a messaging provider cannot return the confirmation Murph needs to track a reply, Murph no longer treats that reply as successfully completed.
Murph retries only when doing so is safe; otherwise it stops instead of risking a duplicate message or attachment.
When a member shares workout summaries with a group, recent connected-device workouts now appear as workout counts and total minutes after sync.
Sharing permissions stay separate: this restores the existing count-and-minutes summary without exposing detailed workouts, routes, heart rate, or private source identifiers.
15 updates shipped in this edition.
New features
Browse practical, research-backed guides for outcomes like sleeping better, lowering resting heart rate, building strength, and training for an event.
Every guide works on its own. All seven goal categories are grouped under clear section labels with no more than 12 goals at a time, and every goal now has a simple visual cue that carries into its guide. Search understands common phrasing such as iron man; when there is no exact guide, Start with Murph opens Messages with your goal already filled in.
A new comparison library explains when Murph replaces, complements, or differs from the health products people already use.
Each guide leads with a practical relationship, then shows fit, tradeoffs, FAQs, current official-source links, and a verification date. The guides are non-affiliate desk research, not hands-on reviews or medical rankings.
Improvements
When your timezone is known, Murph now uses the current local clock for time-sensitive answers instead of treating UTC as your time.
If a trustworthy local timezone is unavailable, Murph keeps the answer in UTC and says local time is unknown rather than guessing.
Larger private dashboards now spend less time preparing saved health data in the background, helping updates finish more reliably.
The data shown and privacy boundaries are unchanged. Foreground messages still take priority, and existing pages continue to read the same saved data.
When Murph uses typed goal save with an existing goal ID, a missing ID now stops the update instead of creating another goal.
Murph can inspect the current goal list, but it will not retry against a different goal without confirmation. This safeguard is limited to typed goal save --id updates; advanced JSON imports keep their existing behavior.
When Murph imports a structured meal, unsupported fields now stop the save instead of being silently ignored.
The recovery result names the supported meal shape without repeating submitted private values. Existing photo and audio field aliases continue to work.
Managed AI usage now reflects the regular Luna model when Luna is selected through Venice, instead of counting the higher Luna Pro rate.
The selected model, reasoning, and reply behavior are unchanged. Murph also includes exact duplicate attachment text once rather than repeatedly, leaving more of your allowance for useful work.
In its weekly health check, Murph can now notice a repeated new behavior tied to something you said matters and send one useful, natural continuation.
A single event, routine shortfall, duplicate note, or poorly timed interruption stays quiet. Without a current intention from you, Murph can only make a neutral observation and cannot create a goal, reminder, or accountability loop.
Replies are more reliable during routine Murph service updates when a newly started runner briefly comes up on the previous release.
Murph still verifies that the runner is on the current release before handling a turn and now makes one bounded replacement attempt when rollout versions briefly differ. If the replacement is also behind or another readiness check fails, the turn still stops safely.
Scheduled reminders now keep their turn while Murph works through a large connected-health backlog.
Background sync retries slow down when they stop making progress, while due reminders and current conversations continue without waiting for that backlog to clear.
After an idle period, Murph now saves its ready-to-go assistant for your current conversation and keeps established conversations ready to continue.
Background work keeps its ordinary startup and cannot take that ready assistant. If the faster path is unavailable, Murph falls back to the durable startup; a full workspace restore still opens with a fresh assistant process.
Reminder saves and edits now stop when schedule details conflict, instead of silently choosing one value and risking the wrong time.
The correction points to the exact schedule field without repeating its contents. Existing one-time, interval, recurring, local-time, time-zone, and device-activity schedules keep working when their fields agree.
Murph keeps private replies focused by bringing in a small set of relevant saved facts and naming the exact target in suitability answers.
Your latest message and the current saved record still win. Background cleanup reads accepted voice-note transcripts like ordinary conversation, leaves newer corrections untouched, and retires a saved fact when your recent conversation makes clear it no longer applies; full detail remains available on demand.
Murph can keep using valid saved knowledge when another saved item needs repair.
Affected reads now say when results are incomplete, preserve clear result counts, and point to a bounded repair check. A damaged requested item is also distinguished from one that does not exist.
Workouts can now be logged from a title, activity type, and duration without requiring a duplicate note.
When both a note and an explicit title are provided, Murph keeps them separate and preserves the chosen title. Existing duration checks still stop incomplete entries before saving.
2 updates shipped in this edition.
Improvements
Murph now keeps delayed wearable imports moving alongside routine background work, helping connected activity and health data reach follow-up work without repeated delay.
Recovery stays bounded so retained imports can catch up without holding up other background work. Interrupted progress checkpoints now converge on the next pass instead of causing repeated retries, and Murph does not mark data processed until the import succeeds or reaches a confirmed failure.
Murph now answers ordinary onboarding and setup questions directly, reserving early progress notes for work with several substantive checks or a longer wait.
Routine context reads and one or two quick calls stay in the main reply, while longer research, document inspection, and other substantial work can still receive one useful update.