Changelog

One saved post can now lead to several Places

Builds 24 and 25 bring the new processing path into use and repair the delivery problems found along the way. A post that recommends several destinations can now return several Places inside one save. Open each Place or add it to a Plan, with its supported dishes and prices kept together.

• Enabled the new extraction and Place-decision flow, with corrections for posts that recommend several venues and for choosing the intended branch
• Improved how named dishes and their original prices are retained and grouped under the Place they belong to
• Restored source thumbnails and repaired result delivery so completed saves do not remain stuck on Finding a Place
• Improved continued processing after manual Place confirmation, while keeping your choices private to your save
• Separated supported Place resolution and thumbnail work from transcript failures, and isolated some provider failures so they do not stop unrelated saves
• Updated private admin diagnostics to show the current Places, dishes, prices and thumbnail, with clearer explanations of partial results

This is not a guarantee that every venue or dish will be found. Some extraction and formatting gaps remain, and dish extraction when a transcript is unavailable still needs work. Recovery for older failed saves, Try again and delete/re-save behaviour remain separate follow-ups.

Next: Make the first save easier for new testers, from installing Wantt and setting up the Share Sheet to returning to a Place.

New rules for identifying Places and Things to try

Build 23 builds and tests a new processing path for posts that mention several Places. The goal is to identify each destination separately and keep dishes, drinks and prices with the Place they belong to. This release is mostly groundwork behind the scenes, with no major visual changes.

• Built support for several independently resolved Places within one saved post, so one unresolved destination does not hold up the others
• Added rules to keep Things to try and their individual prices with the destination supported by the source, leaving uncertain associations unassigned
• Added checks to distinguish an intended venue from its surroundings, such as a restaurant inside a hotel
• Strengthened how later source evidence and personal confirmations are retained without replacing other valid results or another user's choices
• Tested the new path and compared extraction approaches, retaining Jev with the existing xAI fallback for this release

The new automatic processing path is not yet enabled in production. Live saves still use the existing path, and recent device tests showed missed venues, missing prices and fragmented or duplicate recommendations. Build 23 establishes the foundation; it does not yet demonstrate a general improvement in extraction quality.

Next: Use Build 24 to address these gaps and evaluate Luna through the full pipeline before deciding whether to change the extractor.

See saved posts while Wantt finds the Place

Build 22 improves how Wantt identifies Places and picks out Things to try from Instagram posts. Available source images now appear in Recent saves and the saved item before a Place is confirmed, so you can recognise what you saved while Wantt works.

• Improved recognition of venue names and location clues from captions, transcripts and Instagram account names
• Recovered more dishes and drinks explicitly recommended or highlighted in the source post for Things to try
• Added source images before Place confirmation, with a gentle shimmer during processing and a static blurred image when confirmation is needed
• Kept the review flow available when an image is missing, with a clear Image unavailable label
• Expanded internal save history, reporting and image diagnostics to help investigate individual saves

Some image sources still fail to arrive, and difficult posts can still miss a Place or recommendation. This release improves visibility and diagnosis; it does not fix every image-delivery issue.

Next: Improve remaining Place and Things to try gaps, including posts that mention several destinations.

Jev now decides which Place a save refers to

Wantt 0.8.0 changes how a saved post becomes a Place. Source capture, Place identification and useful details now run as separate parts of the enrichment process. Jev makes the bounded Place decision, helping Wantt reach a trusted result much sooner while deeper information can continue to arrive.

• Jev now decides which supported real-world Place a saved post refers to using a faster, bounded path
• Separated source evidence, Place identification and richer details so each can progress independently
• Started retaining the Instagram posting account and explicit coauthor information behind a save
• Expanded internal diagnostics so individual saves are easier to trace and improve
• Removed the transient offline banner while preserving contextual network-error handling

In the Build 21 A/B evaluation, median (p50) Rich extraction took about 0.4 seconds with Jev, versus about 51 seconds on the xAI-first path. That measures the Rich extraction step, not the full journey from save to Place. Place coverage still needs improvement.

Next: recover more venue candidates from difficult posts and improve the fidelity of Things to Try without giving up the speed gained here.

Manage Plans and Place images are repaired

Wantt 0.7.3 is a corrective TestFlight release after yesterday’s 0.7.2 build did not meet on-device acceptance. This combined update focuses on the improvements that now matter for testing: Manage Plans works normally again, and a saved Place remains visually useful when a representative image cannot be captured.

• Repaired the native Manage Plans experience, including its layout and core interactions
• Added a dependable fallback treatment when a Place has no representative image available
• Repaired the fallback-image layout so real source imagery is preferred whenever available
• Strengthened the native iPhone visual and interaction checks used before a build reaches TestFlight

This is deliberately a focused release. The purpose is to make saving, viewing Places and managing Plans dependable again on real devices before asking a wider group of friends to test.

Next: use feedback from physical devices to keep tightening the real-world experience before the next broader release.

Navigation, text size and manual resolving are easier to use

Build 18 is now in TestFlight as Wantt 0.7.1 (build 18). This release brings native liquid-glass controls to the main navigation, fixes layout and control handling at larger system text sizes, makes manual resolving easier to complete, and continues improving how Wantt captures a representative image for a Place.

• Moved the main Places, Plans and Profile navigation into a floating native iPhone liquid-glass treatment
• Added a matching liquid-glass back control to Place detail
• Repaired layout and control handling across core screens at larger Dynamic Type text sizes
• Made manual resolving easier to complete, then return you to the relevant Place or Places list after you confirm a match
• Continued improving representative-image capture while keeping a saved Place useful when a source image is unavailable

Most of this work shows up in the small interactions: navigation stays out of the way, manual help can be completed at the text size you choose, and a saved Place still has value whether or not an image arrives.

Next: keep refining the Place-first experience and the real-world edge cases behind it.

Wantt gets a fresh coat of paint

Build 14 is now in TestFlight as Wantt 0.7.0 (build 17). This is a substantial visual redesign: a darker, more photo-led interface, calmer hierarchy, and a clearer home for the Places and Plans you are building from what you save.

• Introduced a new dark visual theme, with stronger photography and clearer information hierarchy
• Reworked the app around Places, Plans and Profile, with saved source posts retained as supporting history
• Gave Place detail a clearer, photo-led presentation while retaining the Reel as supporting context
• Recognise an already-known Place directly in the Instagram Share Sheet
• Show the Place name, address, open status and things worth trying before a repeat save is confirmed
• Bring saved Reels together under the same Place, rather than leaving useful recommendations scattered across sources

This is not just a reskin. The app is now organised around the real-world Places you might visit and the Plans you might make, while the original post remains there as useful context.

Next: refine this new foundation and keep making the transition from saved post to useful Place feel more dependable.

Places stay visible while details load

Build 13 is now in TestFlight as Wantt 0.6.2 (build 16). Once Wantt has enough evidence for a real-world Place, it can show that Place before every supporting detail has finished processing.

• Show a trusted Place as soon as it is confirmed, rather than waiting for all enrichment to finish
• Keep that Place available if secondary details take longer or cannot be completed
• Preserve the most advanced Place state through backgrounding, refreshing and reopening the app
• Made Instagram Share Sheet saves more reliable when Wantt is not already open
• Fixed note editing so unsaved changes remain available until you explicitly save or discard them

The boundary is deliberately simple: a trusted Place is already useful. Wantt can continue adding details quietly, but it should not take away the place you saved just because one later step is slow or unavailable.

Next: build on this more dependable Place foundation with clearer visual presentation of what you have saved.

Clear saves resolve to Places

Build 12 improves the part of Wantt that turns a saved post into a Place. When the evidence clearly points to one real-world venue, Wantt can now weigh those clues together rather than treating missing optional details as a reason to stop.

• Improved Place resolution for clear, well-supported venue evidence
• Kept genuinely uncertain or conflicting saves in Needs help instead of guessing
• Let two different saves enrich at the same time, reducing the wait when several are saved together
• Repaired Try again for failed saves, with clearer states when a retry is temporarily unavailable
• Strengthened the release safeguards behind enrichment, retries and data changes

This is a more useful middle ground: less unnecessary hesitation when a place is clear, with the same caution when it is not. The app still asks for help rather than assigning a Place it cannot support.

Next: keep improving the remaining real-world edge cases, including messy spellings and bilingual place names, without weakening trust.

Places resolve sooner

Build 11 is a quieter quality-and-speed release, now in TestFlight as Wantt 0.6.0. The aim is simple: when a post clearly points to a real-world place, Wantt should get to the right Place sooner — without turning uncertainty into a confident wrong answer.

• Improved the path from clear venue clues to one useful canonical Place
• Let a trusted primary Place appear before all of the supporting enrichment work has finished
• Kept secondary details working quietly in the background, so they do not hold up the part you came for
• Added a narrow direct validation route for exceptionally clear venue evidence
• Strengthened the checks behind enrichment, including a repeatable set of real-world examples and safeguards against trusted wrong Places

Some of this release is intentionally invisible. The most important improvement is a better path through the middle: save something, let Wantt do the work, and return a Place when the evidence is good enough. Manual confirmation remains there for the cases where it is not.

We also learned something useful. Build 11 improves how Wantt evaluates, protects and delivers the primary result, but the hardest ambiguous posts are not solved by pretending they are clear. That remains the next focused piece of work.

Next: keep improving correct Place resolution where the name is messy, incomplete or slightly different from the real-world venue — while protecting the trust built into the save-to-use loop.

Saving and place resolution are more reliable

Build 10 is a focused reliability-and-enrichment release, now in TestFlight as Wantt 0.5.0. The point is simple: a save should keep moving toward something useful, even when the source or an outside service is less than perfect.

• Rebuilt Plan reordering with a more native iPhone interaction, with order preserved after reopening the app
• Made enrichment more resilient when a provider is slow or temporarily unavailable, with clearer recovery states and a bounded Try again path
• Added Help Wantt find the place when a saved post is taking longer than usual, including likely matches, a short explanation and an easy way to reject the wrong place
• Made user-confirmed places authoritative, while letting other non-conflicting useful details continue to arrive
• Laid the groundwork for trusted repeat saves, structured prices for things to try, and the original date a source was posted

Some of this work is deliberately quiet. A repeat save of the exact same Instagram source can reuse compatible trusted knowledge without bringing across another person’s notes, Plans, visited state or identity. When a post’s original date is available, Wantt can also show its age plainly in the existing Source details rather than pretending every saved thing is new.

There is one accepted visual issue in this TestFlight build: after an imprecise Plan drag, rows can briefly overlap or look stacked. The order itself is correct, persists after relaunch, and the layout returns to normal after reopening the app. We have recorded a focused follow-up rather than hold this release for a non-critical visual defect.

Next: complete the remaining physical TestFlight checks, then use this more dependable foundation to keep making saved things useful.

Saves and Plans get a product-quality pass

Build 8 was the first deliberate product-quality pass across the main Wantt experience. The foundations were already working; this release focused on making the app clearer, more useful and much closer to the product we actually want people to use.

• Renamed Trips to Plans across the user-facing experience
• Redesigned the Saves and Plans overviews for a denser, more content-first library
• Cleaned up primary navigation, back behaviour and screen hierarchy
• Reworked Save Detail so useful place, things-to-try and Plan information leads, with Instagram treated as supporting source content
• Polished Plan Detail across both List and Map views, including a new Wantt map-marker treatment
• Added the updated Wantt app icon to the real TestFlight build
• Established a backend-first production verification process for future releases that change APIs or data models

The release passed its wider physical-iPhone review: existing Saves and Plans survived the upgrade, Share Sheet capture and enrichment continued to work, the new detail hierarchy held up on-device, and the map interactions behaved as intended.

One issue did survive the release. Manual Plan reordering did not work reliably on a physical iPhone. A replacement TestFlight binary — Build 9 — was produced to address it, but the behaviour remained effectively unchanged. Rather than keep burning release cycles on a non-critical bug, v0.4.0 was accepted with Plan reordering recorded as a known issue and deferred to the next build.

Next: fix Plan reordering, then keep improving the reliability and intelligence behind what a save can become.

Wantt gets real user accounts

Build 7 gives Wantt its first proper account foundation. Sign in with Apple is now part of the TestFlight app, while the experience remains local-first and still works without forcing people through a login wall.

• Sign in with Apple is now available in Settings
• Existing Saves and Trips stay intact when you sign in or out
• Anonymous use still works, and signing back in with the same Apple Account returns you to the same Wantt account
• Behind the scenes, account ownership, diagnostics and app-to-server checks were hardened so multiple testers can use Wantt safely and failures are easier to diagnose

Engineering and deployment are complete, and the main production journey has passed on the first physical iPhone: share a new Reel, see it save immediately, enrich it, reopen the app and find the result still there.

Build 7 is conditionally accepted for now. The remaining check is deliberately simple but important: put the same TestFlight build on a second person's iPhone and confirm that the two users' Wantt data stays separate. That test is waiting on Apple's external TestFlight review.

Next: finish the two-user check, then start making Wantt feel much more like the product it wants to become.

One post can contain multiple places and recommendations

The biggest change so far was mostly under the surface. Wantt's original model treated a saved post as something fairly simple, often a single place. Real saves are not that tidy.

We rebuilt the underlying model so one post can contain multiple recommendations and things worth trying, while different shares of the same source can converge on the same canonical content. The app also moved onto a new versioned API with richer place information. That work first shipped as v0.2.0, with the release closeout landing as v0.2.1.

• Support multiple recommendations and things to try from a single saved post
• Added canonical saves and stronger duplicate handling
• Moved the app onto the new /v2 data path
• Added richer place metadata, including opening hours and open-now state
• Kept the previous TestFlight build compatible while the backend changed underneath it
• Completed a release closeout and regression pass before accepting the new release state

Getting there exposed an important distinction between merged and shipped. Some of the new code initially landed while the production backend was still running an older deployment, which left real saves stuck looking for a place match. Redeploying the current backend brought the product back into sync without requiring another TestFlight build.

Next: spend less time rebuilding foundations and more time using Wantt as a real product.

Website

wantt.io is live

The first public home for Wantt is online. It is intentionally simple for now: a place to explain the idea, keep a changelog, and point toward the App Store launch.

• Launched wantt.io on GitHub Pages
• Added the first Wantt visual identity and bookmark mark
• Added light and dark mode styling
• Added Privacy, Terms and Support placeholders ahead of App Store submission

Next: keep iterating on the product in TestFlight and make the public site evolve alongside it.

Milestone

First TestFlight build

Wantt moved from a private web prototype into a working iPhone beta. The core loop now works end to end: save a social post, extract what is useful inside it, and come back to it later in a form that is easier to use.

• Share Instagram posts directly into Wantt from the iOS Share Sheet
• Extract public post metadata and captions
• Use AI to identify places, venues and useful context
• Open extracted places in Google Maps
• Detect duplicate saves
• Remove saved posts
• Support light and dark mode

Next: use it heavily, find the friction, and improve the save-to-use loop before App Store submission.