Say hello to Outa — an assistant that listens, widgets for your home screen, and a stage for stories

Photo: Unsplash

The last phase gave the platform a voice: your own cloned narration voice, a credit system, a marketplace. This phase turns the tables. If OutaStory can speak — can it also listen?

From now on: yes. OutaStory gets Outa, an assistant you can genuinely talk to — out loud, by voice, or quietly, by keyboard. Outa helps when you're stuck writing, explains settings instead of making you hunt for them, and knows the way to support when something's off. It's the biggest thread of these weeks, and it's built — more on this below — with the same care we recently put into the wallet: with clear boundaries, real consent, and not a single byte flowing anywhere without your say-so.

But Outa wasn't alone. In the same weeks, OutaStory moved onto your phone's home screen — with a Story-of-the-Day widget, an audio mini-player, and real app links that open a shared story directly in the app. The home page got a stage: a large Story-of-the-Week presentation with curated category backdrops. Authors gained two long-requested freedoms: deleting stories — with a fair grace period for readers mid-story — and publishing under pen names. And this very blog you're reading is itself one of the new pieces.

On top of that comes perhaps the densest quality wave since the alpha began — and an honest story about a single uppercase letter that took down the entire search. 76 pull requests since the last post, spread across four releases. In order.

Say hello to Outa

The core first: there's now a small, unobtrusive button in the bottom-right corner. One click, one short, clear consent — and then you can talk to OutaStory. Really talk: if you're signed in, you speak with Outa by voice, in natural language, with follow-up questions and interruptions, the way you'd talk to a person. If you'd rather type — or you're on the train — you have the same conversation in text. Under the hood runs the Agents platform from ElevenLabs, the same partner whose voices have carried our audiobooks since the grant; Outa's conversational voice accordingly sounds refreshingly un-robotic.

More important than how Outa speaks is what Outa is allowed to do. We deliberately built the assistant around four roles:

• Writing partner. You're stuck in chapter three, the dialogue feels wooden, the antagonist is flat — Outa knows your story, thinks along, and makes suggestions. But: no suggestion ever lands in your text by itself. Every phrasing is shown to you, and only an explicit "apply" inserts it — the story remains the author's at every second.
• Profile explainer. "Where do I turn off push notifications?" — Outa explains every setting and takes you to the right place. Outa itself may change almost nothing: exactly two things — your bio and your newsletter subscription — and both only after an explicit confirmation in which you see what would change.
• Support guide. Outa answers platform questions from the same help content that powers the support center. And wherever money is involved — a refund, a strange charge — one iron rule applies: Outa never moves money. Instead, Outa pre-fills a support ticket with all the details, and a human decides.
• A tool for the team. For moderation and management, Outa knows the shortcuts — "show me the open tickets" opens the right page. Here too: signposts only, no interventions. Outa links to management surfaces but performs no management actions.

This gradient is the real design: tiered authority. Outa may suggest a lot and explain even more — change almost nothing, and touch money not at all. An assistant that could do everything would have been quicker to build. But nobody entrusts their story to an assistant that writes into it uninvited, and nobody their wallet to one that "takes care of it".

Consent worthy of the name

A voice assistant depends on an external provider — and that's exactly why one point mattered to us more than any feature: until you explicitly agree, not a single byte flows to ElevenLabs. No script is loaded, no connection is opened, nothing. Until your click, the button in the corner is pure decoration — and an automated browser test verifies that on every single build: it watches the network traffic and raises the alarm if even one request heads toward the provider before consent.

Add to that data minimisation in operation: the voice feature is reserved for signed-in users; anonymous visitors can type with Outa — deliberately throttled to a handful of conversations per day and device. The provider key never leaves our server; the browser only receives a short-lived, signed ticket for exactly one conversation. And the privacy policy wasn't "updated soon after" — it was extended with the assistant section in the same release, including data export and a deletion path, as our inventory rule has required for every new data category since the privacy phase.

The assistant as source code

One peculiarity I'm a little proud of: Outa's entire "being" — the personality, the rules, the boundaries, the knowledge about the platform — lives as source code in our repository, versioned like every other line of code. On deployment, the platform automatically reconciles the provider's state with this definition. That sounds technical, but it has a very human consequence: there is no silent drift, no "who changed that in the vendor dashboard?". Every change to what Outa says and may do goes through the same review as every code change — traceable, revertible, documented.

Outa went live with the current release and appears behind a switch, first for the alpha — on the web and in the apps, there including the microphone permission with a clean fallback to text chat if you decline it.

OutaStory moves onto the home screen

Photo: Unsplash

Until now, OutaStory was something you open. These weeks it became something that comes to meet you — right on your phone's home screen.

On Android there are now two home-screen widgets:

• The story of the day. A widget shows a selected story every day, with cover and title — a small shop window between your app icons. Tapping it doesn't open the app's home page for you to click through; it opens exactly that story.
• The audio mini-player. If you've started an audiobook, you see right on the home screen where you left off — with a progress bar and real transport buttons. Keep listening without hunting for the app; and if you haven't started anything, you get a suggestion. So that this stays correct even offline, the device remembers your listening position locally.

The widget taps land precisely inside the story thanks to the second thread: real app links. An OutaStory link — from an email, a chat, a shared post — now opens the installed app at the right place on iOS (Universal Links) and Android (App Links), instead of stranding you in the browser. If the app isn't installed, the right thing happens: the link simply opens the website. As a side effect, this lays the groundwork for domain verification in the app stores — an inconspicuous building block that will matter at public launch.

The first version of the story widget had two teething problems, by the way — covers from the seed catalogue stayed blank, and the daily story didn't populate reliably — both fixed within the same wave.

A stage for stories

Photo: Unsplash

Until now the home page was mostly good at one thing: listing many stories. What it lacked was a place where a single story is allowed to shine. Now it has one: at the very top you're greeted by the Story of the Week — staged large, with cover, description, and a backdrop that fits the story's world. And "fits" means: categories now have curated hero backdrops — two dozen specially selected mood images, which every category inherits as long as no custom one is uploaded. The stage rotates as a gentle carousel, and which story stands on it is driven — like the rest of the home page — by what interests you.

Right below it, listening got its permanent spot: a personalised audio-stories row, reliably in third place, on both variants of the home page. If you use OutaStory mostly with your ears — and since the audio weeks, noticeably more people do — you no longer have to search for the audiobooks.

A shop window of our own: this blog

Until now these posts existed only as files in the engine room. Now they have a home: outastory.com/blog — most likely the very page you're reading this on. The blog runs on Postnomic, is bilingual (every post in German and English), cleanly embedded into the platform — footer, about page, sitemap, search-engine markup — and, as a public shop window, it also survives the coming-soon phase: if you want to know what's being built here, you can read along from now on, before the platform itself opens.

Authors stay in control

Photo: Unsplash

Two freedoms that look small at first glance and fundamental at second:

Deleting stories — with decency. If you want to withdraw one of your stories, you now can. But a platform where people are in the middle of a story must not pull it out from under them from one second to the next. That's why deletion is built as a soft delete with a grace period: the story disappears immediately from the catalogue, search, and recommendations, but anyone already reading or listening may finish it during a transition window — with an honest banner noting that the author has withdrawn it. The author keeps sight of the deleted story in "My stories", and only moderation can remove something immediately and permanently in serious cases. We sharpened this within the first week: the banner wasn't shown everywhere, the "View" button was missing, and your own deleted story still appeared as a recommendation on your own home page — all fixed.

Writing under pen names. Some stories you don't write under the name your colleagues know you by. Authors can now create pseudonyms — each with its own profile, its own address, its own avatar, and its own followers. If you publish under the pen name, nothing publicly leads back to your real account: not the profile page, not the sitemap, not the audiobook. Internally it all remains one account — one inbox, one wallet, one overview. It's the same idea as the voice clone, just reversed: there you give the story your own voice; here you take your own name out when the story needs that.

A lesson: the search that found nothing

Now for the promised honest story. At the start of the month, production search gave the same answer to every query: nothing. No error message, no crash, no red alert — just zero results, for everyone, for everything.

The cause was a single difference in letter casing. The search service stores fields in one spelling (title); our translator in the application code expected another (Title) — and was silently pedantic about it: instead of warning, it simply left every field that didn't match exactly empty. The search index was full, the search worked, the responses came back — and were wordlessly turned into empty shells on the last mile. A bug that crashes would have been found in minutes. A bug that stays silent looks like "well, there just aren't any results".

The fix itself was one line. The real work was the consequence we drew from it: search now reports suspicious silence — if a conspicuous share of queries comes back empty, the observability layer raises the alarm, instead of us depending on chance to notice. It's the same lesson as the quadrupled audio durations a few weeks ago: the most dangerous bugs aren't the loud ones, they're the polite ones. Search has since been verified as restored — and is one watchdog richer.

A second silent candidate fell into the same category: a routine library update had stripped the database connection of its Azure sign-in building blocks — the API simply couldn't reach its database at startup anymore. Here too: pinned down, secured, and a guard check added so that the same trap springs at the next update before it reaches production.

The safety net gets tighter

The quietest and at the same time largest block of these weeks: more than thirty pull requests just for tests. Three things from it are worth telling:

• The production safety net now tests strictly like a human. Our end-to-end tests against the live platform used to sign in through back doors in some places, bypassing the UI. Those shortcuts are gone entirely: every production test now signs in through the real login form, clicks real pages, and asserts on what actually renders — moderation, management, campaigns, payment pages, all of it. What a human can't do through the UI, we don't secretly test through the plumbing — because that gap is exactly where the bugs hide that later hit "only real users".
• Almost every page now has its own test scaffold. In a week-long wave, browser and component tests appeared for the remaining areas — profile, management, moderation, help, contracts, insights. Page coverage was the last red spot on our quality map; it is now visibly shrinking week by week.
• The audit paid for itself immediately. While systematically probing the account endpoints, we noticed that two self-service endpoints accepted a supplied identifier too willingly — a signed-in user could have touched another account's data with crafted requests. The hole is closed; the endpoints now verify without exception that you only operate on your own account. There was no indication of any exploitation — but that's exactly why you probe.

On top of that came follow-ups to last phase's credits: an automatic reconciliation sweep finds stranded narration charges and refunds them without anyone having to file a ticket; the team can now grant credits with a four-eyes trail (every grant lands in an audit log); and new credit packs are purchasable from the moment of deployment, not only after a manual step at the payment provider.

Numbers

• Pull requests: exactly 76 since the last post — the densest block so far. Focus areas: the assistant Outa (four large PRs from foundation to privacy), the Android widgets (seven PRs in plan steps), the home-page stage, app links, the blog, deletion & pen names — and more than thirty test PRs.
• Releases: four — the widgets wave, then stage + blog + app links + audio row, then the repair wave with the restored search, and finally the release that brought Outa live.
• Main themes: a talking assistant with tiered authority and real consent; OutaStory on the home screen (widgets + app links); Story of the Week with curated category stages; a public, bilingual blog; deletion with a grace period; pen names with clean separation; a closed security hole; a search with a watchdog.
• Tests: still green in the high tens of thousands, with the largest single wave of new page and UI tests since the net has existed — and a production safety net that now clicks exclusively like a real person.

What's next

• Hardening Outa in practice — watching the first real conversations, sanding edges, and running the device checklists (microphone, insert, confirmations) on all platforms.
• Widgets onto devices — the Android checklists for the daily story and the mini-player are ready; app-link tap-through tests on iOS and Android follow.
• Privacy sign-offs before public launch — for the assistant as for youth protection, documented review gates from our data protection officer apply before any switches are flipped for the public.
• Store purchases in the sandbox — the real purchase flow on Android and iOS devices is still outstanding.
• Continuing page tests — the last red coverage spot keeps shrinking in weekly waves.

A few months ago, OutaStory was a platform you read. Then one that reads to you. Now it's one that listens — and one that meets you on your home screen instead of waiting to be opened. If you talk to Outa and something feels wrong — an answer that's too forward, a suggestion that feels intrusive, a confirmation missing where you expected one — that is exactly the feedback that counts now. Teaching an assistant boundaries is easier while it's new. Let me know — or tell Outa; after all, finding the way to support is what it was built for.