The first one we kept.
Earliest preserved desktop design. Earlier experiments may exist.
That day ↗BUILD LOG / 08.09.2026 →
Ideas become
real products.
Rosi, Mittelstandkaufen and the other projects: what moved forward, day by day. Source work, observed products and published releases — with measured ChatGPT and Codex activity.
Updated · Notes & recorded metrics
THE FAMILY ALBUM / 12 → 21 SEPTEMBER
Same dog. Quite a journey.
The first design we still have. The little waistcoat. Learning to move and listen. And now a dachshund taking calls above the Alps. Original archive images — the old labels belong to their day.
A little more herself.
Mobile reference; design approved 13 Sep.
That day ↗
Less mascot. More mischief.
The original feedback design. Entrance, roll and dance motions enter production.
That day ↗
Now she has a home.
Archived screenshot of the integrated beta.
That day ↗IN MOTION
She moves. She listens.
Her whole little self, in motion. Preserved listening reference; not a claim it was rendered on this day.
That day ↗
Ready to say hello.
Public entry screen captured 18 Sep.
That day ↗IN MOTION
Taking calls. Mid-flight.
She picks up, listens and answers mid-flight. Silent animation; not a live call.
That day ↗Before 12 September, the visual archive is incomplete. These are the earliest preserved designs, not a claim that every experiment is here. Swipe through the album; tap a picture to see it in full.
THE PROJECTS / DAY BY DAY
What moved forward.
Newest first ↓
The pipeline behind the search.
A new main-branch commit records the crawler and data pipeline that prepare business listings for the platform. The repository now has four recorded commits. This is source evidence, not a claim that every scheduled data run has completed.
GitHub · source verifiedExplore the project ↗
From browsing to a focused acquisition search.
The public site now presents Scout and Autopilot, with search agents, deal analysis and a morning briefing, alongside listing packages for sellers. This observation records the public offer; checkout, email delivery and paid account features were not tested.
Public website checkedExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
A repeatable path for site updates.
Two main-branch commits add scheduled and manually triggered deployment jobs. These document the release workflow; they do not by themselves prove that a deployment succeeded.
GitHub · source verifiedExplore the project ↗
Start with what is on your mind.
Rosi’s opening is being made more personal: follow the person’s actual topic after their name, without a nickname detour. Report handling also distinguishes insufficient conversation facts from missing email verification. Both changes are verified in source; live calls were not tested in this update.
GitHub · source verifiedExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
Vibe coding. Arctic adventures. Ten places.
THEWEEKEND now opens with a native, muted background film and an optional film with sound. The offer is €7,500 net per person: €3,750 at booking and €3,750 thirty days before the start. Applications for 15–18 January 2027 are open to ten selected participants.
THEWEEKEND · video and offer updateExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
Keep the conversation, profile and report connected.
Further source work links accepted conversation details to editable profiles and requested PDF reports, stabilises the voice connection while listening and matches a report to its exact call. Compatible fallback changes are recorded too. This work was committed after the earlier 8 October measurement.
GitHub · source verified · later updateExplore the project ↗
Better timing for the next conversation.
Callback preparation now starts at a safe boundary after the speech response. The checked branches also contain compatible fallback changes. This records committed source work; activation in live calls is not established by this check.
GitHub · source verifiedExplore the project ↗
One search for your next company.
Mittelstandkaufen brings business-sale and succession offers from exchanges, chambers and insolvency sources into one search. The public site shows regional and industry filters, indicative valuations, a market report, search agents and a daily-deal subscription. The website was checked today; email delivery and paid account flows were not tested.
Public website checkedExplore the project ↗
An AI sprint. An Arctic weekend.
THEWEEKEND brings three days of personal AI and entrepreneurship coaching together with Arctic experiences in Swedish Lapland. Applications are open for ten selectively chosen participants; the planned trip is 15–18 January 2027.
THEWEEKEND · initial launchExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
Mittelstandkaufen joins the build log.
The first available GitHub snapshot contains source for search agents, account flows, valuation, market views and deal alerts. It is a code backup, not a complete development history. The single commit is dated 7 October in Berlin; earlier daily activity and releases are not inferred.
GitHub · one code-backup commitExplore the project ↗
Voice and memory, connected more carefully.
Source changes work on the voice connection, a natural ending after silence and profile updates that preserve user edits. Compatible fallback code is included. Product activation remains a separate check.
GitHub · source verifiedExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
A route back into the browser conversation.
The checked branches add first-conversation callbacks in an open browser, alongside a compatible way to pause new callbacks. These commits document implementation and fallback preparation.
GitHub · source verifiedExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
A reply that stays with the conversation.
The source retains chat replies after an automatic profile update and adds push notifications for completed replies. A compatible fallback keeps existing reply readers working.
GitHub · source verifiedExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
Keep the first conversation when you sign in.
New source work preserves the guest conversation through verified account linking and applies existing contact preferences. Further changes prepare automatic profiles and distinguish intentional hangups from connection problems.
GitHub · source verifiedExplore the project ↗
A real practice search, with a clearer view of the market.
Praxiskaufen.com brings listings from 78 sources and 12 specialties into one search. The public site shows price context, personal search agents, market reports and a practice valuation tool. This entry records the observed live product, not a claimed launch date.
Public website checkedExplore the project ↗
The next chapter is business succession.
Handwerkbörse.com applies the same approach to workshops and skilled trades. Its public draft shows business search, listings, market overviews and valuation. It is still labelled as a draft, includes some example data and marks further agent features as planned.
Public draft checkedExplore the project ↗
Two more projects in the workshop.
Praxiskaufen.com and the public Handwerkbörse preview join the existing portfolio with their own branding, actual website views and direct links. The original page design and existing projects remain in place. The development figures are updated separately.
0.8.32Explore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
Memory, call history and an introduction that can finish.
Seven additional commits are recorded on the checked October release branches. The source work covers persistent profile memory, confirmed call exchanges in the temporary chat and first-introduction timing with a completed goodbye. These are verified source milestones; this entry does not assert that each change is active in a live call.
October release branches · source verifiedExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
The local development record continues.
Available local Codex records resume on 1 October for the Rosi and diemer.ai workspaces. Only usage records are aggregated: unique response IDs, input and output tokens, and distinct owner chats. Conversation content remains private. Missing usage between snapshots is not estimated.
Local Codex record · measurement resumesExplore the project ↗
Measured development activity ＋
—recorded tokens
—GitHub commits
—documented releases
—code lines added
PersonalityTimingNext
Good timing is part of having a personality.
Last night’s work gave Rosi more rules for reading the room. This morning added a very sensible one: a good-morning message should arrive in the morning. Small details; quite a lot of code behind them.
- Prepared source · 28 SepPace, answer length and humour can adapt to the conversation. “Be serious” wins immediately. Rosi keeps her character; the register changes.
- Prepared source · 29 SepGood-morning messages stop being an option at 11:00. Permission, quiet hours and “stop” remain firm boundaries; silence means less contact, not more pressure.
- Next · not active yetConnect these modules to the runtime and an explicitly opted-in contact flow. No automated messages are being announced. The last verified Live release remains the direct call entry from 27 September.
Technical notes Build changes, checks & numbers ＋
2,496,244tokens burned
2GitHub commits
0documented releases
—code lines added
Good morning. Before eleven.
The prepared contact module now rules out a good-morning message after 11:00. Consent, quiet hours and an explicit stop take priority over learned preferences. This is merged source, not an active messaging service.
Contact timing · preparedExplore the project ↗
ConversationMemoryNext
A proper conversation. Still the goal.
The latest live release makes reaching Rosi simpler. Now the useful work is in the details: the pause before an answer, a goodbye that finishes, and memory that works with your permission.
- Latest verified live release · 27 SepDirect call entry and ten measured minutes per Berlin day. Today’s data refresh preserves that release; it does not announce a new rollout.
- Dev / preparedThe retry-close fix is on Dev. Further runtime cleanup and capacity work remain prepared until their own Live activation and checks.
- NextVerify iPhone farewells and response timing, finish the email time bonus and prove the account-memory handover. Personality gets better through a conversation that actually works.
Technical notes Build changes, checks & numbers ＋
90,323,699tokens burned
6GitHub commits
0documented releases
—code lines added
Same Rosi. A better ear for the moment.
A prepared conversation module adjusts pace, length and humour from conversational cues. “Shorter” or “be serious” takes priority immediately; one laugh does not rewrite her personality. The runtime connection is still pending.
Conversation register · preparedExplore the project ↗
A little hello. Never a guilt trip.
The next contact idea is prepared in code: only with permission, less often after silence, and a stop that stays a stop. Messages would draw on the same daily story as calls. Sending, channels and scheduling still need integration and approval.
Considerate contact · preparedExplore the project ↗
The next step is a conversation that flows.
The latest verified release is the direct-call entry from 27 September. Remaining work includes measured response speed, a fully completed farewell on iPhone, the email time bonus and account-memory acceptance. A retry-close fix is on Dev; its Live activation is still separate. Today’s note updates the outlook, not a newly deployed feature.
Current outlook / no new release claimExplore the project ↗
CallsNativePlatform
Open the page. There she is.
One less gate between you and the dog. The call button is right there now. Her story still matters, but it does not have to stand between you and saying hello.
- LiveThe EN/DE entry now goes straight to Call Rosi. No automatic Story overlay or Ready gate before the call button.
- Prepared, not liveThe next capacity package passed scoped tests and CI. The wider capacity is not activated and has no simultaneous-call acceptance yet.
- Still being checkedNative Build24 availability was reported. Actual iPhone farewell completion, the email bonus and account memory remain separate open checks.
Technical notes Build changes, checks & numbers ＋
195,592,420tokens burned
1GitHub commits
1documented releases
—code lines added
Less entrance. More Rosi.
The English and German landing pages now show Call Rosi directly, without an automatic Story overlay or Ready gate. Installed files, public pages and the visible desktop entry were checked. This frontend release did not change the backend or prove a new device call.
R238 / direct call entrySee the live result ↗
Room for more conversations. In the workshop.
The next capacity package has passed its scoped offline checks and remote CI. It is prepared source, not activated Dev or Live capacity, and not evidence of ten simultaneous media calls. The active backend remains the separately verified earlier runtime.
R237 / next capacity stepExplore the project ↗
VoiceStevieConversation
More time together. And a guest appearance.
Ten minutes with Rosi, split into as many calls as you like. Stevie has a better cue for his entrance. And behind the scenes, the work is about making the conversation feel less like waiting for software.
- LiveTen measured minutes per Berlin day replace the five-call limit. Stevie is scheduled once after sixty conversation seconds, when the conversation has room for him.
- Live, with boundariesEarlier speech generation, the approved aftercall screen and separately optional technical analytics are connected. That does not prove every reply is fast or the iPhone farewell is complete.
- Still in the workshopThe email time bonus and account-memory acceptance are unfinished. The retry-close fix reached Dev; Live needs its own activation and checks.
Technical notes Build changes, checks & numbers ＋
661,278,700tokens burned
10GitHub commits
4documented releases
—code lines added
Less waiting is the point.
Validated reply text can enter speech generation earlier, alongside focused prompt and turn-boundary fixes. Source tests and the deployed runtime were checked. Early timing samples were too small and uneven for a reliable speed-gain claim; fluid real conversations remain the goal.
R227 / response pathSee the live result ↗
A guest with a sense of timing.
Stevie is scheduled once per logical call after sixty seconds of confirmed conversation, at the next eligible pause. Reconnects should not create another entrance. The timing change reached Dev and Live with seven scoped tests; a new physical-device or complete audio performance was not claimed.
R230 / Stevie timingSee the live result ↗
Say goodbye. Then tell us what worked.
The approved aftercall screen is connected to the existing account flow, with a checked skip path. Technical web analytics require a separate opt-in and exclude conversation content. Feedback, verified identity and long-term memory remain distinct; account memory was not newly proven by this UI release.
R231 / aftercall and optional analyticsSee the live result ↗
Ten minutes. Your choice of calls.
The live allowance now follows ten measured conversation minutes per Berlin day, without the earlier five-call limit. Confirmed setup failures and unmeasured legacy estimates do not consume talk time. The additional ten-minute email bonus is still unfinished; it must not be advertised as active.
R235 / measured talk timeSee the live result ↗
StoryCallsPersonality
More Rosi. Fewer hoops before hello.
Her story now opens the call in German or English. Five calls a day, no compulsory account. Underneath, the work is still very real: help her listen, reply sooner and treat the camera as context, not the whole conversation.
- LiveThe final EN/DE intro films are connected to the call entry. Five public calls per Berlin day and a fix for the duplicate native intro remove a few more hoops before hello.
- Live, still being checkedCamera context now takes second place to the conversation. The service is open again after a deployment interruption. This is a shipped change, not a claim that reply speed or every iPhone call is solved.
- NextQuicker replies, a simple voice-first entry with optional video, and reliable feedback handover. New movement transitions and broader native testing are still in the workshop.
Technical notes Build changes, checks & numbers ＋
1,266,469,851tokens burned
17GitHub commits
8documented releases
—code lines added
A little more time together.
The live runtime replaced the two-minute per-call ceiling with ten measured minutes across the Berlin day. The qualified Stevie sequence was included. The controlled Dev run did not establish a complete audible farewell or physical iPhone acceptance; those remained separate checks.
R225 / daily conversationSee the live result ↗
Her story. In both languages.
The final German and English intro films are now connected to the public call entry. The selected language preloads, playback starts muted and sound is an explicit choice. Public files and browser playback were checked. Older embedded preview films are separate work; this does not claim every story player has switched.
R212 / final story introsSee the live result ↗
Five chances to say hello.
The public call allowance is now five calls per Berlin calendar day, without requiring a login. Confirmed setup failures before the greeting is acknowledged do not consume a call. Policy and deployed source checks passed; this is not proof of every device or conversation.
R214 / public callsSee the live result ↗
One intro is quite enough.
The public runtime now leaves the intro to the native app when that app owns it, preventing the duplicate introduction. Focused fixtures and public file checks passed. A separate short simulator call showed 278 video frames and a started greeting; physical-phone listening and multi-turn latency remain separate checks. External TestFlight rollout is not announced.
R215 / native introSee the live result ↗
A conversation, not a running camera commentary.
The live backend now treats automatic camera images as background context and explicitly prioritises the conversation. Six synthetic checks passed. That is a shipped instruction change, not a measured speed or empathy improvement. Following a deployment interruption, service was reopened and its public bootstrap checked at 12:11 UTC. Real-call responsiveness, feedback delivery and optional video entry still need work.
R216 / conversation firstSee the live result ↗
The phone finally shows her moving.
A focused native simulator run presented 246 video frames and recorded a completed cached greeting. The decoder fix is on the public site. This proves that specific short video path, not a physical iPhone listening test, a long conversation or a complete dynamic answer.
R209 / native videoSee the live result ↗
You can just listen. She can handle that.
The live call can keep running in its consented listen-only mode when microphone access is denied or revoked. Startup now keeps old controls behind the white loader, and waiting/error notices are clearer. One controlled Dev video call completed its greeting and a typed reply; that reply still took about 8.5 seconds. Speed and native account admission remain open work.
R210 / listen-only and call readinessSee the live result ↗
A pause should not become a goodbye.
When already-playing audio is unexpectedly paused, the existing resume control now reaches that same player and call. The public update also removes inherited zoom that clipped the white 2D character. Activation and exact file checks passed; 20 synthetic audio cases and visible desktop/mobile framing checks support this scope. Real-device pause recovery and a full new call were not re-tested.
R211 / same-call audio recoverySee the live result ↗
PersonalityCallsPublic release
A little dog. A very real public release.
Today the work moved onto the public call pages. Less ceremony before hello. More of her character in the conversation. And a fairly long list of little things that still needed fixing.
- LiveThe approved call entry and personality reached the public runtime: one consent step, a checked intro and the right order when saying goodbye and leaving a rating.
- LiveYou can type with the microphone muted and keep the scene you chose. German personality bindings were corrected; a complete audible Bavarian conversation still needed checking.
- Still in the workshopThe Stevie encounter stayed in controlled Dev. Native playback, camera joins and call startup still needed focused work; a published package is not a promise that every call is perfect.
Technical notes Build changes, checks & numbers ＋
1,430,342,686tokens burned
21GitHub commits
2documented releases
—code lines added
One hello. Fewer hoops.
The public release brought the approved single, initially unchecked consent box, intro-readiness checks and leaving the call before the rating screen. Dev and production activation and public file checks were recorded. The native call entry has focused module checks; this is not a new physical-phone or listening acceptance.
R202 / public call entrySee the live result ↗
Still her. Even when you type.
The approved German personality was joined to the public runtime. Typed replies retain the selected scene when the microphone is muted. The live release was activated with no active calls and its files and entry endpoints checked. A complete audible Bavarian conversation remained unverified; the Stevie encounter was still a controlled Dev feature.
R203 / personality and quiet microphoneSee the live result ↗
Let her arrive. Then let her be Rosi.
The current work targets the opening: the approved logo loads first, the entrance is prepared, Rosi walks in, and only then do the surrounding controls appear. A scoped fix avoids relying on Safari preload alone and releases the controls even if the clip after the entrance fails. Another prepared change lets approved quiet movements begin after 2.5 seconds while keeping speech, privacy and hidden-page safeguards. These changes passed local source tests; the current handoff still requires actual browser testing and publication. They are not part of the running Dev release yet.
In progress / entrance and quiet momentsExplore the project ↗
ConversationMovementApp
Same Rosi. From hello to goodbye.
I want to call Rosi and feel she is there. Today that means fewer awkward starts, one consistent little dog on screen, and a conversation worth staying for. Quite a few small fixes. One fairly big ambition.
- On DevLater Dev updates made the mobile Kitz pair about 78% smaller and joined cached restarts, frame continuity and wind/cowbell cues. Two intro-to-walk cycles were observed; this is not a full call or physical iPhone acceptance. Production stays unchanged.
- Browser checkedA later neutral Main test received a complete relevant answer; Sebastian also reported that video worked and partly flowed. Earlier answer failures and Safari issues remain part of the record. This is not measured latency or complete call acceptance.
- In the workshopI have installed and tried the existing TestFlight app; the newer build is still separate. Stevie’s new transition passed its visual check. Sound and the movement afterwards remain open.
- NextLess waiting for a proper answer. More natural eyes, head and tail. Keep her personality; make spending time with her feel effortless. That is the aim, not a finished claim.
Technical notes Build changes, checks & numbers ＋
2,174,213,793tokens burned
21GitHub commits
0documented releases
—code lines added
A lighter backpack for the little dog.
The mobile Kitz walking and speaking pair now totals 6.66 MB instead of 30.8 MB: about 78% less data, with the original frames and timing retained. Dev delivery, seeking and caching were verified. A phone-size browser check buffered both clips before a call without opening the microphone or camera. That is delivery evidence, not a physical iPhone or complete conversation test.
Dev / smaller mobile videoExplore the project ↗
A second hello. A little more of her world.
The next Dev update joined cached call restarts, frame continuity, handling of a hidden 2D view and approved wind and distant cowbells. Main observed two consecutive intro-to-walk cycles, including a restart with retained cache. The optional camera preview can be moved within its bounds. Actual voice, every motion join and physical iPhone behaviour still need separate acceptance; the later requested entrance sequence remains unfinished. Production is unchanged.
Dev / movement and surroundingsExplore the project ↗
One dog. From the first frame onwards.
The development build now includes a visible first frame, clearer consent controls, an immediately accessible end-call action and stricter ownership of the selected scene. The latest follow-up adds same-scene response handling, a camera cue and softer excited delivery; its hosted files were verified at 15:16 UTC. Kitz appeared in the latest real Dev video call, but the answer failed to complete. A full call and its response speed are not yet accepted. Production is unchanged.
Dev / video entry and continuityExplore the project ↗
The app gets the same foundations.
Native presentation, profile handling, chat and feedback connections have joined the development work, and the verified release was handed to the existing app team. Sebastian has installed and tried the existing TestFlight build. The newer app build matching the latest Dev release is still a separate delivery; that handover is not proof it is already on the phone.
Dev / app connectionsExplore the project ↗
Less waiting. More being there.
The aim is a Rosi you can actually spend time with: quicker meaningful replies, natural eyes, head and tail, and reactions that fit what she is doing. The newer deployment still does not establish a working complete call or its speed. The new Stevie transition has passed its visual review; original sound, the following movement and the actual call integration remain open. The existing personality stays; a new gag does not get to jump the queue ahead of a good conversation.
Next / faster replies and natural movementExplore the project ↗
A quick hello is only the beginning.
The next Dev build brings together earlier speech recognition, conversation context and voice timing work. Its running code and frontend were verified. In the recorded check, a cached greeting started promptly and the normal question got a relevant answer, but full answers still took about 6.3 and 9.1 seconds by browser callbacks. That is too slow; the cause is being narrowed down. These are technical timings, not an acoustic measurement. The public production conversation remains unchanged.
Dev / reply timingExplore the project ↗
The right look. Now the right transition.
A gaze take and a Kitzbühel walking take have received visual approval. Connecting them to real listening, speaking and interruptions is a separate step: the gaze runtime and complete hiking scene still have open checks. A nice clip is not automatically a working conversation. The new walking scene is not being advertised as available to call.
Movement / look and walking reviewExplore the project ↗
MovementPersonalityStory
Looking alive is the easy bit.
The real Rosi never waits for a feature release to be herself. Her digital counterpart needs a bit more work: listen properly, move naturally, and have something of her own to say. That is today’s workshop.
- In the workshopMore life in the Selfie scenes: listening, speaking and staying in the same place throughout a call. The hiking loop still needs a correction. Looking almost right is not quite Rosi.
- In the workshopYour inner dog, with an explanation: a playful profile idea based on what you choose to share and confirm. With researched breed traits, a little humour and the option to disagree.
- In the workshopHer Story player now passes a local browser check: it starts quietly on scroll, with German and English sound controls. Mobile layout checked too. The finished English film and public activation are separate steps.
- On DevThe later Mac test completed a spoken question, a typed follow-up and the feedback handover. The test message was stored and its notification arrived. Public release and iPhone acceptance are separate steps.
Technical notes Build changes, checks & numbers ＋
2,111,088,140tokens burned
10GitHub commits
0documented releases
—code lines added
A conversation, then a proper handover.
The later Dev check completed a spoken question, a relevant answer and a typed follow-up. Rosi introduced herself once. Rating, a short recorded test message and the move to the sharing screen worked; the test feedback was stored and its notification reached the inbox. This was a Mac browser test of the development build. It does not establish physical iPhone performance, word-level lip sync or a new public release.
Dev / voice and feedbackExplore the project ↗
The story now plays through its local check.
The existing Story film now decodes and starts muted on scroll in the local browser check. German and English sound controls, keyboard operation and a 390-pixel mobile layout were checked. A simulated call-start signal stops the film and its download. This is player integration evidence, not a completed English film, an acoustic review or a public release.
Story / native player checkExplore the project ↗
A new Dev checkpoint, with real calls still under review.
The development backend was updated and its running commit verified on 22 September. Call start, two typed voice replies and durable feedback were observed. A missing follow-up reply and missing display of a typed user message still needed correction. The public conversation frontend stayed unchanged; this is not a claim that the complete next release is ready.
Dev / conversation checkpointExplore the project ↗
A little movement. A lot of timing.
The next scene work connects listening and speaking in Kitzbühel, Pilates and Munich. The selected place should stay with Rosi throughout the call. Native playback checks have exposed a too-short repeating action in the hiking scene, so it stays held for correction. Small idle reactions also need clean entries, exits and interruptions before activation. These are local integration and media checks, not a new public scene release.
Motion / listening and speakingExplore the project ↗
Your inner dog needs a reason.
Rosi’s next optional profile feature connects things you have actually shared and approved with sourced breed traits, then explains the playful resemblance in German or English. Owning a Chihuahua does not make you one. Weak evidence should produce no match. The local contribution passed 39 synthetic checks, including correction, deletion and account separation; real model, voice and hosted acceptance remain open. This is a playful analogy, not a personality diagnosis.
Personality / your dog alter egoExplore the project ↗
Her story gets a quieter entrance.
The existing Story film is being joined into RosiGPT’s next homepage build with loading on scroll, muted playback and a clear sound control. Starting a call stops the film and its download. Seven focused integration checks passed locally; mobile, audio and hosted acceptance remain ahead. The current interim film is retained. This does not announce the finished English film.
Story / player integrationExplore the project ↗
LocationsPersonalityStory
A dachshund with places to be.
The next step is a Rosi with a little life of her own. Places, friends, opinions — and the same recognisable character wherever you meet her.
- In the workshopMunich, hiking in Kitzbühel, her Lambo, Pilates and Käfer: the first Selfie settings we’re bringing together. Each needs its own look, believable movement and a little life in the background.
- In the workshopOne personality with friends, humour, opinions and daily stories that line up across her conversations and her world. Private chats stay private.
- Already on screenRosi GPT’s Wingsuit, Harley and home scenes are already visible. On DIEMER.ai, she now really lifts the phone to her ear in the silent Wingsuit clip. Scene-aware calls and the finished English story film remain separate work in progress.
- On DevConversation, movement, profile and Story came together in the development backend. The complete frontend and call experience still need their shared acceptance checks.
Technical notes Build changes, checks & numbers ＋
2,223,585,339tokens burned
10GitHub commits
1documented releases
—code lines added
More of the pieces meet on Dev.
The cumulative conversation, movement, profile and Story backend reached the development environment on 21 September. The combined frontend stayed a separate preview while account, camera, voice and full-call checks continued. Public conversation entrypoints remained unchanged. A Dev deployment is progress, not the completed public release.
Dev / conversation, profile and StoryExplore the project ↗
Yes, she actually picks up.
The new Wingsuit clip has returned: Rosi checks her phone, brings it to her ear, listens and answers while flying. A silent animation for the website; not a live call or a newly voiced story film.
Phone pickup / render reviewExplore the project ↗
A world she can actually react to.
We are connecting what really appears in the scene with what Rosi knows about it. A background moment should be visible before she comments on it; a serious conversation should not be interrupted by a gag. Scene, sound and conversation checks are still part of the next candidate, not a claim that every location is live.
Selfie / scene-aware conversationExplore the project ↗
Five places. One recognisable Rosi.
The next Selfie set brings together Munich, Kitzbühel hiking, her Lambo, Pilates and Käfer. The work combines location-specific appearance, movement and background activity with the same character. These are the initial scenes being integrated and qualified; the complete scene catalogue is not publicly released.
Selfie / first scene setExplore the project ↗
One Rosi, with a life of her own.
The cumulative personality now defines one consistent Rosi: warm, cheeky, independent, with friends, opinions and small imperfections. Her authored daily episodes are designed to stay coherent across conversations and Social, using the same episode/version. Private conversations do not become Social content. This is a documented product contract and integration work, not a claim that the full runtime or Social publishing is already live.
Personality / own-life canonExplore the project ↗
A little more of her world, already on screen.
The public rosi.dog page shows her existing Wingsuit flight and other parts of her own-life world. The phone overlay there is a labelled demo, not a recorded phone-taking gesture. The current Story behind Rosi film remains available while its replacement is being finished. This website milestone is separate from the conversation runtime.
Rosi’s own-life pageSee the live result ↗
A blink. An eye-roll. A little more Rosi.
The real Rosi can change my mood without saying a word. A look, a ridiculous little reaction, and I’m gone. That is today’s brief for her AI counterpart: more life in her face, more of her own character in the conversation, and Selfie adventures that belong to the moment. The next motion and German/English personality build is prepared; ten new scene clips have rendered. Now comes the part I obsess over: does it actually feel like her? Full conversation, movement and sound checks are still ahead. The public conversation build stays on R70f1.
21 SEP / ROSI’S DIARYExplore the project ↗
feat(motion): a look, a blink, a very Rosi eye-roll.
More expression is the job: where she looks, how she listens, and the occasional deeply unimpressed eye-roll. The next motion build connects these cues to the conversation. Source checks have passed; natural movement, mouth timing and the complete call still need their real-world check.
R76 / motion workshopExplore the project ↗
feat(personality): more of her. In both languages.
Less generic assistant. More of the original dachshund. The next cumulative German/English personality is pinned and checked at source level. It still needs to earn its place in a real conversation. R75 with the previous personality remains the last verified Dev baseline; R76 is being prepared.
R75 DEV / R3 personality candidateExplore the project ↗
feat(selfie): the conversation catches up with the scenery.
The next build connects the active Selfie scene to the right conversation and rejects events left over from an earlier one. More locations are coming, with Kitzbühel first in the planned selection. Ten new scene clips have rendered; visual, sound and full-call checks come before release.
R76 / Selfie integrationExplore the project ↗
MotionSelfieStory
An eye-roll is a feature now.
Rosi doesn’t spend a whole conversation in one pose. A glance, a little distraction, a suspiciously well-timed reaction — that is the work today.
- In the workshopMore gestures, softer returns to listening and two rendered squirrel reactions for Main and Selfie. Their picture and transition checks still come before release.
- On DevNew Selfie scenes, the Wursti voice and opt-in memory work. The story film is being refined with our real adventures.
- LiveThe feedback screen has a working way back to the conversation. Small fix. Much nicer ending.
Technical notes Build changes, checks & numbers ＋
1,854,761,588tokens burned
16GitHub commits
1documented releases
—code lines added
story: our real adventures get their own film.
One original dachshund. A lot of adventures together. The new story-film cut has been activated on the Story behind Rosi page. Dev playback passed; the final public playback check is still open. The main conversation release stays separate.
Story film / final playback checkExplore the project ↗
feat(motion): more than one way to be Rosi.
Today’s mission: less statue, more dachshund. We’re bringing the qualified motions into one catalog and working on more variety while she speaks. Each clip has to return naturally to listening. This next motion package is still in the workshop.
Motion / R75 workshopExplore the project ↗
fix(feedback): an easy way back to the conversation.
The post-call feedback screen now has a working way out on the public beta and on Dev. Dev also clears the visible chat for the next call, separately from any memory you explicitly choose to keep. A small fix that makes the whole flow feel less sticky.
R70f1 + feedback fixSee the live result ↗
feat(dev): Selfie, Wursti and a better feedback handover.
R74 is active on Dev, with new Selfie clips and a checked Realtime 2.1 call using Wursti. Ratings and voice feedback reached their stored receipt and sharing step in the test. Feedback-email summaries still need work; the Mini voice path stays on Dev after an audio failure. The public conversation build is still R70f1.
R74 DEVExplore the project ↗
feat(motion): less statue. More dachshund.
The real Rosi never sits in one pose for an entire conversation. Fair point. Today we’re giving her more ways to lean in, react and settle back into listening. A second sustained main pose is on Dev; smoother returns and more Selfie variety are still being checked.
Motion / today’s focusExplore the project ↗
feat(expression): a squirrel has entered the conversation.
A quick distraction, a curious lean, a cheeky look. We’re building small reactions that fit what is happening, then let you keep talking. Two new squirrel-reaction clips are rendered for Main and Selfie. Their visual and transition checks come before release.
Main + Selfie / in progressExplore the project ↗
story: one original. Quite a big reason to build.
She comes everywhere with me. Everyone wants a Rosi. There is only one original — which is a wonderfully unreasonable reason to build an AI dachshund. The story page is on Dev, and we’re refining its film with our real adventures. The new cut is still in review.
Story behind Rosi / in progressExplore the project ↗
feat(dev): a place for the things worth remembering.
A user area, explicit opt-in memories and a way back to the same Rosi after email confirmation. R72 also brings the alternate main pose and work on the Wursti voice. These changes are on Dev; the complete memory, camera and voice-feedback checks are still open.
R72 DEVExplore the project ↗
SelfieVoiceMobile
The Selfie chapter begins.
Suddenly it gets a little more face-to-face. The public beta gained its Selfie chapter, while the next ideas kept growing in the workshop.
- LiveSelfie conversation work reached the public beta, alongside corrected German and English branding. Voice checks covered interruption, follow-up and a complete goodbye.
- On DevA faster camera entry, larger Selfie view and another listening gesture. The user area, lasting memory and extra poses remained in development.
Technical notes Build changes, checks & numbers ＋
2,128,474,200tokens burned
55GitHub commits
2documented releases
0code lines added
fix(selfie): the camera catches up with the conversation.
The camera view appears immediately, the entry runs more smoothly, and a new listening gesture joins the Dev build. Email confirmation leads back to Rosi. More natural motion is being added carefully, with the return to listening getting as much attention as the reaction itself.
R71f DEVExplore the project ↗
release: the Selfie chapter reaches the public beta.
The public beta is now on R70f1, with the Selfie conversation work and corrected German/English branding. The newer user area, persistent-memory work and extra pose variants remain on Dev. One live product; one workshop where the next version earns its place.
chore(status): current Rosi progress and build-token history
The diary records the public R64f release, R70b/R70c Dev checkpoints and remaining launch work. Local build-token aggregates run through 19 September.
v0.8.18See the live result ↗
fix(selfie): the conversation keeps going after camera frames
The narrow R70c server fix processed eleven replies, including the camera frames, and saved the rating on Dev. The voice-feedback send still needs its complete end-to-end finish, so this is a verified Dev checkpoint — not a live-release claim.
R70c DEVExplore the project ↗
feat(dev): a cleaner first screen, larger Selfie and restored rating
R70b is active on Dev with the direct How entry, a larger Selfie stage, three reachable mobile controls and the original rating graphic restored. The next layout pass removes mobile chat and makes the Selfie call fill the visible screen; the new motion sequence stays out until its head transition is smooth.
R70b DEVExplore the project ↗
release(voice): Teddy, a shorter entry and steadier conversation
R64f is live with the Teddy voice, the short approved entry, qualified motion and a conversation-continuity fix. Hosted voice QA confirmed one greeting, two recognised inputs, interruption and follow-up, one complete farewell and provider hang-up. This was browser QA, not a physical iPhone test.
MobileVoiceFirst hello
Less ceremony. More conversation.
Getting to Rosi should be simple. Today was for the first few seconds: a clearer start, a bigger dog and fewer things in the way.
- LiveA shorter greeting, clearer microphone permission and a larger Rosi on mobile. Call controls stay reachable.
- LiveThe iPhone audio check was corrected and connection errors became clearer. One greeting follows permission.
Technical notes Build changes, checks & numbers ＋
1,696,541,081tokens burned
189GitHub commits
4documented releases
8,721code lines added
feat(site): Space-Rosi stays with you, until her pitch desk takes over
Faster 6/7, original vested Rosi in the flagship project, and pitch buttons below both illustrations. The transparent Space-Rosi stays at the viewport edge, fades out only while the pitch section is visible, and returns through the footer. Reduced-motion and mobile legal-link access are preserved.
v0.8.17See the live result ↗
feat(site): the real Rosi, her build diary and a quieter first screen
Speaker control inside the film, captions without a badge and tighter spacing. A personal Rosi story with original photographs, a new seated ideas-desk portrait and the original speaking videos on the pitch page. The homepage feed and default diary now focus on Rosi; other projects stay in the archive.
v0.8.16See the live result ↗
feat(site): Space-Rosi, permanent captions and a fresher build diary
A new Space-Rosi costume for the pitch, the existing voice and recording flow, a refined sound control and always-on subtitles. Public project versions refresh with dated checks. Cross-project release notes and a Project progress entry sit together. Privacy and Imprint offer English and German.
v0.8.15See the live result ↗
feat(site): personal AI / Human, dedicated pitch, independent project overview
Independent DIEMER design with ROSI GPT and DEEPDIEMER as the primary projects. AI / Human explains why curiosity, impatience and precise feedback fit this workflow; four of the ten learnings are revised. Pitch to Rosi moves to its own page with a larger character and the existing verified-email recording flow. The external ROSI diary dependency is retired.
v0.8.14See the live result ↗
fix(iphone): recognise the supported audio path
The audio capability check now recognises the iPhone streaming interface. Connection failures are reported more clearly. This patch keeps the current conversation design and personality.
feat(rosi): the vision, the story and the timeline
Personal AI takes the lead: ROSI for company, DEEPDIEMER for self-reflection. Three ROSI chapters cover vision, personal story and daily progress. RobinPrompt and the stealth Prediction Market sit in the footer.
v0.8.13See the live result ↗
fix(consent): one greeting after permission
Consent and microphone permission now come before one short greeting. Start and How controls stay locked while a conversation connects or runs.
fix(mobile): larger Rosi, clearer microphone steps
A larger character, simpler entry and clearer consent targets. Speaking-loop transitions and listening gestures have been refined.
fix(film): single-line captions and a cleaner toolbar
Removed the film overline. Published a versioned, single-line English caption track and right-aligned the CC and sound controls. Film and audio are unchanged.
v0.8.12See the live result ↗
fix(navigation): learnings replace the retired engine room
Removed the outdated Engine Room and its project/landing-page teasers. My learnings is now a main-navigation destination. Old Engine Room URLs redirect to the ten learnings.
v0.8.11See the live result ↗
feat(entry): a shorter greeting, clearer start
A shorter welcome, consent at microphone connection and more reliable HD animation selection. Call controls remain reachable on mobile.
fix(audio): slightly deeper film narration
Lowered the existing film mix by another half semitone locally. Video packets, captions and timings retained. Audible autoplay is attempted where browsers permit it; the existing Sound on control handles iPhone playback restrictions.
v0.8.10See the live result ↗
fix(media): quieter opening, speech-synchronised Rosi
Removed the opening obsession line from voice and captions. Original video packets and all following text and timing are retained. Rosi’s briefing and contact handover now use speech-synchronised animation with the unchanged voice in a single media stream; reduced-motion and audio fallback included.
v0.8.9See the live result ↗
feat(hero): widescreen film, captions and sound control
Complete personal montage retained in 16:9. Timed English captions, a gently deeper soundtrack, touch-sized audio controls, and adaptive 360p/720p/1080p delivery.
v0.8.8See the live result ↗
VoicePersonality
Same dachshund. Different moods.
Some days call for a gentle voice. Others could use a little nonsense. Rosi is learning to sound less one-size-fits-all.
- LiveGentle, playful and emotional voice choices, with a fuller personality and movement connected to the conversation state.
Technical notes Build changes, checks & numbers ＋
702,245,281tokens burned
59GitHub commits
1documented releases
18,653code lines added
fix(media): restore the new personal film with a targeted edit
New personal montage retained. Pilot face zoom removed; desk transition repaired. Adaptive delivery and the single sound toggle remain.
v0.8.7See the live result ↗
feat(voice): three tones, fuller personality
Gentle, playful and emotional voice choices. Conversation-state motion and a revised German personality core make the delivery more consistent.
fix(media): widescreen edit, sound state; confirmed legal operator
Wide-format film without the pilot close-up or inconsistent desk shot. Chapter controls removed; sound button reflects browser playback. Imprint and privacy controller updated to Unicorn Asset Management GmbH with the confirmed ROSI GPT details.
v0.8.6See the live result ↗
feat(media): adaptive clean film; privacy and responsive polish
Clean 44-second story with adaptive 360/720/1080 delivery. Clear legal footer, refreshed privacy notice and minimal consent records; full portrait framing and responsive layout checks.
v0.8.5See the live result ↗
perf(security): lighter films, fewer blocking requests, HTTPS policy
Smaller desktop/mobile animation and handwriting font, a versioned CSS bundle, early poster loading and static compression. HTTPS HSTS added. Voice and mail processing remain on the verified backend.
v0.8.4See the live result ↗
feat(learnings): ten notes from the current AI cockpit
Dated personal learnings on creativity, microfeedback and agent structure. Dedicated page, landing teaser and accessible chapter navigation. No changes to voice processing or mail transport.
v0.8.3See the live result ↗
fix(mobile): fit the action to the phone
The Magic Prompt action fits the usable phone viewport. Answer targets stay round; optional Rosi audio sits below the animation.
feat(projects, mail): focus, stealth & responsive Rosi emails
ROSI GPT focus badge, non-clickable prediction-market concept and mobile pitch CTA. Branded HTML/plain-text emails; actual editor mail reached Inbox. Authenticated SMTP setup remains pending.
v0.8.2See the live result ↗
ui(pitch): one frame, clearer start & agent handovers
Compact laptop pitch frame; one personal story section; agent workflow overview below the release log. Voice capture verified locally. Authenticated mail setup remains pending.
v0.8.1See the live result ↗
fix(submissions): open the form reliably
Read requests are sequenced and temporary loading failures can be retried. 94 editorial prompts. Submissions enter a private review queue; public community ranking is still in progress.
Beta 0.1.12See the live result ↗
VoiceMotion
A little less waiting. A little more Rosi.
Speaking, listening, reacting: the little transitions matter. Today the public beta got closer to a conversation that flows.
- LiveNew speaking and listening movement, personality refinements and a quicker microphone start. A transition fix keeps the outgoing frame visible until the next clip is ready.
- ReviewedFour motion renders were reviewed. Coffee and roll scenes still had integration conditions; rejected takes stayed out.
Technical notes Build changes, checks & numbers ＋
572,652,533tokens burned
62GitHub commits
2documented releases
12,009code lines added
feat(pitch): inline capture, transcript review & new layout
30-second capture, email verification and editable transcript. New opening screen, project KPIs, varied Rosi motion and manifesto artwork. Authenticated SMTP setup and production inbox verification remain open.
v0.8.0See the live result ↗
fix(motion): hold the outgoing frame
Keep the last frame until the next clip is ready. Restore the approved ear cue; retain the Start-created audio context for captions. Release acceptance remains open.
V39.7 / R56See the live result ↗
fix(film): remove the pause control
Remove the extra animation control across public pages. Keep chapter selection and reduced-motion handling. Deployed bytes verified; owner copy preserved.
v0.7.1See the live result ↗
media: four renders, 604 frames reviewed
Coffee and roll clips carry integration conditions; rejected takes remain excluded. Media review is complete for the recorded package. Main integration and live transitions are still pending.
Motion packageExplore the project ↗
feat(catalog): 20 new prompts, EN / DE
86 published prompts; six missing language counterparts added. ChatGPT / Claude variants. Idempotent import preserves prompt IDs, URLs and existing votes.
Beta 0.1.9See the live result ↗
feat(funnel): test-first entry & provider routing
Remember the selected AI tool for the browser session; full-text copy fallback. Six legacy routes redirect with HTTP 301. 380 deployed files matched their source hashes.
Rosi asks the next question. I can edit my own words.
A short voice conversation turns a pitch into something clearer. The owner editor lets me click, rewrite, preview and publish the site’s text. The build log now puts real daily activity next to the work, and the story is a little more me.
v0.7.0See the live result ↗
The blank prompt has fewer excuses now.
66 editorial prompts, three beginner collections and clearer starting steps are live. Self-reflection, everyday help and lifestyle: find a useful starting point, copy it, and make it yours.
Less explaining what I might build. More showing what I am.
A shorter, English-only site puts current projects first. Swipe through the projects, see what is next, or let Rosi explain how to pitch me an idea. The story now includes the mistakes and the work behind the results.
v0.6.0See the live result ↗
Rosi has a little more to say. And show.
The V39.7 beta is live with new speaking and listening motion, personality refinements and a quicker microphone start. It is another step toward a more natural conversation. The full experience is still in progress. Public beta access is now open.
V39.7 / R55See the live result ↗
A party hat. A better match. Small things matter.
The WhatsApp dachshund now wears a party hat. Find your Rosi has a cleaner group illustration, and Find your human gets its own black-and-tan companion. Three little details, now live.
DesignVoicePersonality
The pieces start working together.
The approved design meets the actual product. Then come the less glamorous bits: getting into a call on an iPhone, connecting her personality and making the tiny joins work.
- Published that dayThree documented activations: the approved layout, a better iPhone entry and personality 1.3.17. The old release records survived in the design archive.
Technical notes Build changes, checks & numbers ＋
1,001,568,203tokens burned
64GitHub commits
3documented releases
22,334code lines added
An idea, a spaceship, and a dachshund on board.
The site is live with the workshop-to-orbit film, a voice inbox for your ideas and my story in two languages. The manifesto sounds more like me now. Rosi contributed her thoughts on the snack budget.
v0.5.3See the live result ↗
Good prompts deserve to get passed around.
RobinPrompt’s public beta brings 43 editorial prompts together. Find something useful, copy the complete prompt and make it your own. Guest votes stay pending while account verification is being finished.
Public betaSee the live result ↗
Less waiting. More getting to the good bit.
The welcome, personality-test buttons and result motion got a polish. The homepage now reveals its text without waiting for the video, and the two tool buttons remain within reach on smaller laptop screens.
v142–145See the live result ↗
Looked aligned. Until the tablet test.
Longer descriptions pushed neighbouring project screenshots out of line at tablet width. Shared layout rows fixed it. The useful lesson: a tidy desktop screenshot is not evidence that the layout works everywhere.
A real lesson from the buildRead the build story ↗
The approved design becomes the real interface.
The approved desktop and mobile layout was integrated and verified on the host at 13:35 CEST. Recovered from the original 15 September activation record.
Final design integrationSee the live result ↗
A better way into an iPhone conversation.
The iPhone conversation entry was revised and activated at 14:01 CEST. This is the historical beta release, not the current product version.
Personality joins the conversation.
Personality 1.3.17 was connected in the beta and activated at 14:42 CEST. The dated record describes this integration, not completion of all planned features.
MotionCharacter
An entrance. A roll. Possibly a dance.
The real Rosi has never needed a slide deck to make an entrance. Her AI counterpart is getting a few moves of her own.
- In the workshopEntrance, gentle roll and a small dance entered motion production. The finished movement still had to prove itself.
Technical notes Build changes, checks & numbers ＋
606,840,234tokens burned
50GitHub commits
0documented releases
10,308code lines added
A little more life in the little dog.
The approved Rosi character is getting an entrance, a gentle roll and a small dance. New motion scenes entered production; this day’s record is a work-in-progress checkpoint, not a live-release claim.
V39.4 motionExplore the project ↗
DesignMobile
She looks like herself. Keep it that way.
The look got its approval. Now the challenge is to make the little dog feel just as right when you actually talk to her.
- Design approvedA shared mobile and desktop reference, with the important call controls within easy reach on a phone.
Technical notes Build changes, checks & numbers ＋
—tokens burned
109GitHub commits
0documented releases
762code lines added
This is Rosi. Now let’s keep her Rosi.
The mobile and desktop design received its final approval. Voice, character and animation keep moving forward around that reference. On a phone, the important call controls need to stay comfortably in view.
Design approvedExplore the project ↗
VoicePersonalityAvatar
First, the parts have to become one dog.
A voice on its own isn’t Rosi. Neither is a nice picture. Today was about getting the pieces to work together.
- In the workshopShared connections for voice, personality and avatar: the groundwork for listening, replying and moving as one character.
Technical notes Build changes, checks & numbers ＋
—tokens burned
34GitHub commits
0documented releases
2,386code lines added
Teaching the parts to talk to each other.
The voice, character and avatar components received shared interfaces. It is groundwork for Rosi to listen, respond and move as one character. A building milestone, with the finished experience still ahead.
FoundationExplore the project ↗
Even the little buttons needed the right plumbing.
An earlier ChatGPT work session caught a measurement problem: new calls to action were not being recorded correctly, and internal navigation could look like a new visit. The fix and checks were prepared for the main build. This is a documented work item, not proof it shipped that day.
ChatGPT build historyExplore the project ↗
Technical notes Build changes, checks & numbers ＋
—tokens burned
0GitHub commits
0documented releases
—code lines added
No dated project notes in the available archive yet.
Technical notes Build changes, checks & numbers ＋
—tokens burned
0GitHub commits
0documented releases
—code lines added
No dated project notes in the available archive yet.
Technical notes Build changes, checks & numbers ＋
—tokens burned
0GitHub commits
0documented releases
—code lines added
No dated project notes in the available archive yet.
Technical notes Build changes, checks & numbers ＋
—tokens burned
0GitHub commits
0documented releases
—code lines added
No dated project notes in the available archive yet.
Curious about the build numbers? Tokens, GitHub commits and build history ＋
THE WORK, IN NUMBERS
Little steps. Quite a trail.
Code lines added—Recorded additions · through 19 Sep
Active project chats—Historical total · through 25 Sep
Tokens burned—
GitHub commits—Rosi + Mittelstandkaufen · checked branch history
Calendar days · Europe/BerlinToday is still in progress
What is counted. What is still missing.＋
Updated 14 October 2026. Published history through 8 October is preserved. New GitHub milestones and available local Codex usage are added after that measurement. Dates use Europe/Berlin.
- Tokens
- Recorded input plus output, deduplicated by response ID. Cached input is already part of input; reasoning is already part of output. Agent usage is included. The new increment contains 7,790 response records and 1,207,866,784 tokens from available local Rosi and diemer.ai sessions. No new Mittelstandkaufen usage was found in the checked local records. Other providers, missing periods and unrecorded work are not estimated. Tokens are not spending, revenue or completed features.
- Chats
- The historical chart still ends on 25 September. The latest-period figure counts 11 distinct owner chats active after the 8 October snapshot. Agent chats are excluded from that count. It is not added to historical chat totals because conversations can overlap.
- GitHub commits
- 763 previously recorded commits plus 11 unique Rosi commits across 46 checked branches and 3 Mittelstandkaufen main commits: 777 in total, including 773 for Rosi and 4 for Mittelstandkaufen. Rosi main and iOS main contain no new commits in this period. This is partial branch history, not a count of live features. Fallback and rollback commits count as development work.
- Mittelstandkaufen
- The first recorded commit is a code backup dated 7 October in Berlin. The new commits document a deployment workflow and the crawler/data pipeline. A complete earlier daily development history cannot be reconstructed. The public offer was checked on 14 October; no successful payment or email delivery is inferred.
- Code additions
- Historical measurements through 19 September are retained. Unmeasured later days remain blank; a source snapshot is not counted as newly written lines.
- Privacy
- Only aggregate metrics and edited product notes are public. No chat text, account details, response IDs or private repository contents are published.
NEXT ENTRY: YOUR IDEA?
Rosi is taking notes.
I’ve got room for another “what if”. She’s got very good ears.
Tell us what to build ↗