Your workspace grew folders — and apps that open into it. The shared space your apps read and write is now a real filesystem. Make folders, move things between them, and browse it properly. MyLLM can do all of it too, so you can simply ask it to tidy up. • Files, with folders — create a folder, open it, and move a file or a whole folder into it. Swipe any row to move, rename or delete. Deleting a folder that still holds files asks first, and tells you how many. • Ask MyLLM to organise — "put the shopping lists under lists/ and archive last year's notes" now has somewhere real to go. • Open a file with an app — hand any document straight to an app that can read it. • Workbench, new in the App Gallery — the app your files open into. It renders Markdown properly, CSV as a real table, JSON, logs and plain text. Choose the typeface, size, line spacing and column width, switch on Paper mode for long reading, and use the formatting toolbar when you edit. Save straight back, or save a copy anywhere in your workspace. • Live app tiles — an app can now show one thing on its own icon: a streak, a count, a score. The tile turns over to show it, then turns back. • Swipe to your apps — swipe in from the right edge of a chat to open MyLLMos, and back from the left edge to return. The handle above the keyboard still works exactly as before. • A Tools screen — see everything MyLLM can do, grouped by what it's for: which tools are switched off, which ones you actually use, and what each one takes. • Python can open your files — hand real files to Python and it can read them normally, then write its results back where your apps can pick them up. • Command Center — start with real monitors instead of a demo, keep a history of what changed, and remove anything from the dashboard by touching and holding it. Fixed • Copying a long answer could freeze the app and leave the message menu stuck until you force-quit. Long copies now stay on this device; everyday copies still reach your Mac and iPad. • Files landed where you expected. MyLLM could save into a private area only it can see and still tell you the file was saved — it now knows the difference, and says so plainly if it happens. • Connected tool servers work properly. If you connect a server with a lot of tools, most of them never reached the assistant. Tools you connected or wrote yourself now come first, and the assistant can find them when it goes looking. Everything here stays on your device.

What's New in MyLLM 5.0 Your apps have been saving things for months. Now you can ask about them and be told when they change. • Ask your files, not their filenames This runs entirely on your phone. Nothing is uploaded to search your own documents. • Automations that notice An automation used to run when you started it, or at a time you picked. It can now run when something changes. You can now finds the right file by what's in it, not by matching words in its name. Ask about the hotel booking you noted somewhere and it finds that instead. The answer comes back with the file it used, so you can check it. Tell MyLLM "whenever anything under vocab/ changes, summarise the new words" and it sets up a watch. Next time one of your apps saves a file there, the assistant reads it and tells you what's new, without you asking. • Ask MyLLM about a file, in one tap Open anything in the shared workspace and there's now an Ask MyLLM menu: summarise it, explain what it contains, turn it into flashcards, or find other files about the same subject. It hands the file to the chat and answers there, so you don't have to describe a document you're already looking at. • Quick start, your way The shortcuts on an empty chat could be added and removed with a long-press but nothing on screen said so. There's now a Customise button. Drag to reorder, swipe to remove, and a switch to hide the whole block if you'd rather start every chat on a clean screen. Removing a built-in shortcut hides it; removing one you added deletes it; and Restore brings the built-ins back while keeping the ones you made yourself. • A console for the apps you build The in-app editor could already write an app and show you a live preview, but if the finished app misbehaved there was nowhere to look. Turn on App Gallery → gear → Developer → Show app console, and any app you open shows what it printed and any errors it hit, at the bottom of its own screen. It's off by default, because a debug strip on a working app is just clutter. • Setting up with a cloud key actually works now If you use a hosted model with your own API key, onboarding had no route for you. It offered an on-device model or a server on your computer, and the second option quietly switched you back to a local server even if you'd chosen otherwise. There's now a proper "Use a cloud model with my own API key" option, with the same providers, the same key field, and the same reminder that your messages leave the device. • Local models were being offered half their tools A quieter one, found while testing. MyLLM decides how many tools to offer the assistant based on how much context your model can hold. If it didn't recognise the model's name, which is true of essentially every custom or self-quantised local model, it fell back to assuming a small context and cut the tool list roughly in half, while at the same time sending the model a context twice that size. It now uses the context size you configured. If you run your own model under your own name, the assistant can reach noticeably more of what it should have had all along. • Connect to MCP servers that use their own auth header MyLLM has been able to add remote MCP servers for a while, and it always sent your token the standard way, in the Authorization header. Most servers want exactly that. A few reserve Authorization for something else and define their own, Google's newly released Sovereign Agent Mesh, for instance, expects X-Sam-Authentication, because the standard header is already spoken for by the sign-in sitting in front of it. There's now an optional "Auth header" field when you add or edit a server. Leave it blank and nothing changes. Fill it in and your token goes out under that name instead. If the connection is ever redirected to a different host, the token is stripped from both header names rather than being handed to wherever the redirect points. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.5.8 Finding a model you actually want, in a list of four hundred. • Search the model list If you connect MyLLM to OpenRouter, the model list runs to well over four hundred entries — and until now it was a scroll wheel of raw identifiers like "meta-llama/llama-3.3-70b-instruct". Finding anything meant spinning past hundreds of names you didn't want. A user wrote in about it, and they were right. The model list is now searchable. Type "sonnet", or "gemini flash", or just the maker's name, and the list narrows as you go. Models are grouped by who makes them, so you can also just browse to the section you want. Every model now shows its proper name rather than its identifier, along with how much context it can hold. • Free models are marked Some models cost nothing to run. They're now labelled in the list, and there's a switch to show only those, so you can find them without cross-referencing a website. Models whose price we can't confirm are simply left unmarked — an unlabelled model means "we don't know", never "it's free", because guessing wrong there is the kind of mistake that shows up on someone's bill. • Cloud models can finally use their full context Underneath that list was a quieter problem. When MyLLM fetched the available models from a provider, it kept only their names and assumed every single one could hold about 32,000 words' worth of context — regardless of what the model actually supports. So a model built to read 200,000 was being treated as though it could manage a sixth of that. That's fixed: MyLLM now reads each model's real limit. In practice that means longer conversations before older messages start being trimmed, and noticeably better answers about long documents, since more of the document fits in one go. • The version shown in Settings is the right one Settings → About compares the version you're running against the one on the App Store, and for some people those two numbers disagreed even when they were perfectly up to date. It turned out Apple serves that information from several caches which don't always agree with each other, and MyLLM was trusting whichever answered first. It now checks all of them and takes the newest. It also remembers what it found, so the line no longer goes half-blank after you reopen the app, and it's labelled "App Store" rather than "Latest" — because if you're on a test build, being ahead of the store is normal, not a fault. • Your saved documents are stored more carefully An internal safeguard around how the Documents library is written to disk. Nothing you'll see, but the library can no longer be dropped if the write is interrupted. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.5.7 Big documents actually get read now — all of them, including the scanned ones. • Attach a long PDF and it reads the whole thing Until now, MyLLM only pulled the first stretch of text out of a document you attached — and it never told you. Attach a long report and the assistant would answer confidently from the opening pages as if it had read the lot, with no hint that the rest had been dropped on the floor. That was our bug, and it's fixed. MyLLM now takes in around ten times as much text from a document. When a file is still bigger than that, it doesn't throw the rest away any more. The whole document is indexed on your phone, and the assistant looks things up in it as it answers, so a question about page 300 gets an answer from page 300. Excerpts come back with page numbers, so you can check it. • It tells you how much it read Attach a file and you'll now see exactly what went in underneath the filename, how many pages the document has, how many were read, and how many needed the extra work below. If anything was left out, it says so in orange instead of staying quiet about it. No more guessing whether the assistant saw the part you cared about. • Scanned documents work A PDF that's really a stack of photographs, anything you scanned, photographed, or were sent as a scan, had no text for MyLLM to find, so it came back empty with no explanation. MyLLM now reads the words off the page itself, on the device, and folds them in with everything else. No network, no account, nothing sent anywhere. Pages it genuinely can't make out are counted and reported rather than silently skipped. • Faster, and no longer freezes on a big file Reading a large document used to lock the app up while it worked. It now happens in the background with a progress indicator, so you can keep typing. Importing a long document is also dramatically quicker, and a long-standing fault that filled your document library with hundreds of junk fragments — quietly crowding out real results when the assistant searched your files — is gone. Everything here happens on your phone. Searching your own documents uses on-device text understanding, and reading scanned pages uses your phone's own text recognition — neither sends your documents anywhere. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.5.6 Apps that can follow a moving phone, a Speedometer that admits when it's lost the signal, and a gallery that survives a bad connection. • Speedometer is better Go through a tunnel, step onto a train, or start it up indoors, and it used to just sit there — a confident number on screen, nothing moving, no way to tell whether it was working or stuck. Now it says what's happening: a pulsing "searching" while it hunts for the first fix, "reacquiring" when the signal drops, and a plain dash instead of a speed it can no longer stand behind. It picks the signal back up on its own, and the fault that could quietly wedge it shut until you closed and reopened it is gone. • Your apps can follow a moving phone properly Behind that is a new ability any MyLLMos app can use: instead of asking "where am I?" over and over, an app can open one live location session and be told as you move. That's the difference between guessing your speed from how far you've travelled and reading it off the GPS chip directly — so speedometers, run trackers and live maps start quicker, hold steadier, and recover better when the signal comes and goes. Apps can also ask for a precise fix when they need one, instead of the rough-and-cheap one that suits a weather lookup. It's still off until you say otherwise (Settings → MyLLMos Apps → "Allow apps to use your location"), iOS still asks you as well, and it's still only ever while the app is open — closing it hands the GPS straight back, because a location left running is the fastest way to flatten a battery. • The gallery copes with a bad connection If the servers hosting the app catalogue are slow or unreachable, the gallery used to go blank or fill with empty tiles. It now falls back to a second host for everything — the list, the artwork, and the apps themselves — and if both are unreachable it shows you the catalogue you last had, with a note saying how old it is, rather than nothing at all. • Fewer tools failing for no good reason When your assistant reaches for a tool — a web search, a file, an email — it has to hand over the details in exactly the right shape, and different models and servers disagree about what that shape is. MyLLM was strict about a couple of those disagreements, and the result was a tool that refused to run and an error message blaming the wrong thing. It now understands every form they come in, so searches that used to fail with "missing required parameter" simply work. And when something genuinely is missing, the message says what was actually received instead of only what wasn't — the difference between the assistant fixing its own mistake and repeating it. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.5.5 - your assistant is choosier about the tools it reaches for. • Sharper tool picking Every time you ask for something, MyLLM decides which of its abilities to offer the model. It was being far too generous — a message with "read" or "save" anywhere in it handed over all thirteen file tools, and asking about your battery handed over eight. Now each question brings up only what it actually needs: "what's my battery like" offers one instead of eight. Fewer options to wade through means the right one gets picked more often, and replies start sooner. Nothing has been taken away — anything not offered up front is still found the moment it's needed. • Kinder to smaller models If you run your own model, this is the one that matters. MyLLM now works out how much room your model actually has and sizes the list to fit, instead of sending the same long list to everything. Smaller local models used to spend their attention reading through options and lose track of what you'd asked; now they get a list they can comfortably hold. • Fixed: patent tools appearing where they didn't belong Under certain circumstances, MyLLM was quietly offering it a set of tools built for one specific server — which then failed if the model tried to use them. It now checks properly, so those tools stay where they belong. As always: private and on-device by default — your data never leaves your phone.

Your posts start moving, and your steps finally sit where you can read them. Social Studio makes reels now Any post you design can become a short animated reel: pick how it moves (Drift, Pop, Slide or Pulse), choose how the words arrive — including word-by-word, the way captions land in the reels you already watch — set the length, and watch it play right there before you commit. Save it and you get a real video in your Photos at 1080 × 1920, ready for Reels, Stories or TikTok, with your caption and hashtags already copied. Reels are silent on purpose, so you add the music where you post them. Your apps can make video Behind that is a new ability any MyLLMos app can use: turn what it draws into a real video and save it to your Photos. Animated posts, a timelapse of a drawing, a chart that builds itself in front of you. It's all rendered on your phone — nothing is uploaded, and iOS asks you before anything is added to your library. Stride — a step tracker that's actually readable A new app in the gallery, built around one idea: you shouldn't have to decode a chart to find out how you're doing. One big ring, one very large number, and a plain sentence that just tells you — "you've walked 6,450 steps, about 4.7 km, 1,550 to go — roughly 16 minutes of walking." Your last seven days sit underneath as tappable bars, and your goal is set in plain language rather than numbers alone: Gentle, Steady, Active or Ambitious. Three text sizes, kilometres or miles, and a goal streak that keeps counting. Your apps can read your step count Behind Stride is a new ability any MyLLMos app can use: your daily steps, distance and the last seven days, read straight from the iPhone's motion chip. It opens showing a real week — no waiting a day for it to fill up. It's off until you say otherwise (Settings → MyLLMos Apps → "Allow apps to read your steps"), and iOS asks you as well the first time. The numbers are read on your phone and go nowhere else — no account, no upload, no fitness service in the middle. As always: private and on-device by default — your data never leaves your phone.

A calmer, steadier chat. Scrolling that stays put — your message settles at the top of the screen and the reply writes in below; no chasing, no jumps, no blank moments, plus the little arrow when a long answer runs past the screen. Update badges that show up on time — MyLLMos home-screen badges appear immediately, even fresh after launch or offline, instead of waiting on a gallery visit.

What's New in MyLLM 4.5.2 Your apps now tell you when they've grown. App updates, right on your home screen When an app in your MyLLMos gets a new version, a small badge now appears on its tile — no more digging through the App Gallery to find out. Long-press the tile and tap Update Now, or use the new Update All banner at the top of your home screen to bring everything current in one go.

What's New in MyLLM 4.5.1 Your own photos, in your apps. • Use any photo as a background Mini-apps can now open your camera or photo library directly, so "Use a photo" just works. Try it in Social Studio: drop in your own photo, pinch to zoom, and drag to position it behind your post. Any app that needs a picture from you can use it now. • Gallery updates land faster When we publish an app update, it now reaches your App Gallery straight away instead of trickling out over the following hours. Already live for everyone in the App Gallery (no update needed): Social Studio's new photo backgrounds, with pinch-to-zoom. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.5 Your assistant reaches further — and reports back. • Command Center A new MyLLMos app that watches the things you care about — your servers, sites and data feeds — on one live dashboard. Point it at any URL that returns JSON and it tracks the numbers for you. Then just ask the assistant "how's everything doing?" and it reads the room across everything you're monitoring. • Agent results that finally read well When an automation finishes, its results are now properly formatted — clear headings, links you can tap, and relevant images pulled from the pages it found — instead of a wall of raw text. Search results are cleaner too: no more half-finished snippets or leaked page code. • Scheduled agents remember their settings Set a daily "Research Topic" and it now keeps the topic you gave it, so an unattended morning run always knows what to look up. Enter it once; every run reuses it. • Email that writes back Ask MyLLM to reply and it can now send email from your account — or draft replies for you to review and send yourself. Build an auto-responder automation, like an out-of-office, with built-in safeguards against reply loops and no-reply addresses. Sending uses secure SSL/TLS and works with Gmail, Yahoo and most providers. • MyLLMos, your way Reach your saved apps from a button in the top bar instead of the pull-up handle — or have MyLLMos open automatically the moment you launch. Both are optional, in Settings. And a new 4-apps-per-row layout fits more on screen. Meanwhile in the App Gallery (live for everyone already, no update needed): the Welcome tour now shows off Skills, Agentic and Command Center, and there's a new Slovak lessons app, "Petra", teaching the basics with the same flashcard system as our other language apps. As always: private and on-device by default — your data never leaves your phone.

New in 4.4: Skills and Agentic join your apps on the MyLLMos home screen — install reusable know-how for the assistant, and set up automations that run on a schedule. The assistant can now grow its own skills, hand tasks to your apps, and offer to use what you've already installed. All on-device.

What's New in MyLLM 4.3 Your assistant can think twice, run your Shortcuts, and speak in character. • Deep Think A new brain button next to Send. Tap it and your next question gets the slow, careful treatment: the model works the problem in stages, checks its own reasoning for holes, and only then answers. For the questions where being right matters more than being fast — tricky bugs, big decisions, homework you'll be quizzed on. • Your Shortcuts, on voice command Ask MyLLM to run any of your Apple Shortcuts — "run my Good Night scene", "log this with my Water shortcut" — and it does, passing text along if needed. If the shortcut replies, the result lands right back in the chat so the conversation can carry on with it. Your automations just became part of the dialogue. • Branch a conversation Long-press any message and Branch from Here now creates a proper fork: the new chat remembers where it came from, shows a "Branched from…" breadcrumb you can tap to jump back, and keeps its own copies of any photos. Explore a tangent without losing the thread. • Every persona gets a voice Long-press a persona to give it its own speaking voice — your Writing Coach no longer has to sound like your Code Reviewer. Read Aloud and Walking Mode pick the right voice automatically. • Walking Mode, less awkward silence While the model composes a reply, you now hear a soft, gentle pulse instead of dead air — the thinking sound (switchable off in Settings → Voice). Small change, surprisingly big difference on a walk. • Dictation in the language you meant Apps that listen in another language (like Talk Tutor's French) now reliably get a recogniser for that language — and the recorder shows which language it's listening in, so there's never a mystery about why "J'habite à..." came out in English. • Sturdier app editing When the assistant edits one of your saved mini-apps, ambiguous changes are now refused instead of applied everywhere at once, and failed edits explain exactly what didn't match. Fewer surprises, better patches. Meanwhile in the App Gallery (live for everyone already, no update needed): Mochi the talking pet now goes for walks through a scrolling meadow with butterfly friends and dreams while it sleeps, Brick Studio lets you and the AI build brick models together — complete with a parts list to gather from your real brick box — and Talk Tutor's tutors found their proper voices. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.2 Your assistant's memory got instant, and your apps got beautiful. • Recall any conversation, instantly Ask about something from weeks ago — "what did we decide about the trip?" — and MyLLM now finds it in an instant, across your entire chat history, not just recent conversations. Both exact phrases and rough descriptions work. Everything stays on your phone: the search index is built on-device, never backed up anywhere, automatically strips anything that looks like a password or key, and can be switched off (and wiped) in Settings → Memory. • 333 app icons to choose from The icon picker for your MyLLMos apps grew from 24 art icons to 333 — eight packs of glossy 3D icons covering tools, games, food, animals, nature, travel and more, based on Microsoft's open-source Fluent Emoji. Your home screen deserved better than a symbol on a gradient. • Apps can listen in their own language Mini-apps that take dictation can now tell MyLLM which language you'll be speaking — so a Spanish tutor can actually understand Spanish. Watch the App Gallery: Talk Tutor is about to put this to good use. • Files that actually delete Fixed: deleting a file in the MyLLMos Files section could fail with "file not found" while the file stubbornly stayed put. Deletions now work every time, clean up empty folders behind them, and the underlying bug that occasionally created empty 0 KB files is gone too. • Seven new apps in the gallery Since the last update: Talk Tutor (a private language partner for Spanish, French and German — it corrects you gently and nobody hears you practise), Comic Box (chat with the original 1996 Comic Chat characters and watch every exchange become a comic strip you can save), Internet Radio, Image Converter (HEIC → JPG at last), Air Quality, Birthday Book and Decibel Meter. Pull down the springboard and tap App Gallery — they're live for everyone right now. • Small print, now findable The gallery's terms and content policy are now linked right from the App Gallery, for the curious. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.1 Smoother, quieter, and your apps can hear you now. • Buttery streaming, even in long conversations Replies used to stutter as they streamed into a long chat — the longer the conversation, the worse it got. We rebuilt how streamed text reaches the screen: the transcript now sits still while only the growing reply updates. Hundred-message conversations scroll like new ones, and reading history mid-reply no longer fights you. • No more invisible ceilings on long replies Big generations — a complete HTML app, a long document — used to be silently cut short, and the model would keep "rebuilding from scratch" into the same invisible wall. Max response length now defaults to No limit (new option in Settings → Performance Tuning) for server and on-device models alike, on-device context windows are 2–4× larger, and when a reply does hit a genuine limit MyLLM notices and has the model continue where it left off, stitching the parts together. • Stop means stop Tapping Stop during a long on-device reply could take ages to actually stop — the model had to finish digesting the whole conversation first. It now checks for your tap continuously, so generation halts in under a second, keeping whatever had already streamed. • Microphone support for MyLLMos apps — fixed Apps that listen — the new Guitar Tuner, Voice Memos — couldn't actually reach the microphone, even with the permission switched on. Two deep issues are fixed: apps now run in a secure context (so the mic API exists at all), and the audio session switches to a recording-capable mode the moment an app is granted the mic — with sound kept on the loud speaker, where it belongs. Tune your guitar to the cent; your recordings record. • Tighter "Internet off" for apps If you switch app internet access off, it's now enforced at the browser level too: no images, no scripts, no requests of any kind. Off means off. • Four new apps in the gallery Subscription Tracker (what does it all add up to?), Packing Planner (checks your trip's forecast, builds your list), Sky Tonight (tonight's moon, planets and the Space Station — real astronomy, computed on your phone) and Guitar Tuner (cent-accurate, five tunings, metronome included). Pull down the springboard and tap App Gallery — they're live for everyone right now, no update needed. There's also a fuller welcome tour for anyone new you show the app to. • New ways to keep MyLLM The free first month stays. After it, choose what suits you: £0.99/month, £9.99/year, or £19.99 once for lifetime — every option unlocks everything, no tiers, no ads. Already bought MyLLM, either as a paid download or with the one-time unlock? You're unlocked for life — nothing changes for you. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 4.0 - now with Hermes 3 • It learns your workflows After finishing a multi-step task, the assistant can offer to save the procedure as a reusable skill — next time, it just knows how. You approve every save, and every learned skill can be reviewed, disabled or deleted in Settings → Skills. Skills use the open Agent Skills format, so skills written for other AI tools install here too, including one-tap install links. • It remembers past chats Ask "what did we decide last week?" — the assistant can now search your previous conversations and quote the relevant part. The search runs entirely on your iPhone; your chat history never leaves the device. • Daily automations Any automation can now run on a daily schedule — a morning briefing, a news digest, whatever you set up. iOS grants background time on its own clock, so timing is best-effort; MyLLM always catches up when you open the app, and results arrive as notifications. • Hermes 3 — the biggest on-device model yet Nous Research's Hermes 3 (3B) joins the built-in model catalog — the most capable model MyLLM has ever run fully offline. iPhone 15 Pro or newer recommended. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 3.9 MyLLM is now free to try — and it works out of the box. • Try everything free for a month MyLLM is now a free download with every feature included for your first month. Keep it forever with a single one-time unlock — no subscription, no ads, and every future update included. Already bought MyLLM? You're unlocked for life automatically — nothing to do, nothing to pay. • A fresh, simpler start The new setup gets you chatting in under a minute. On supported iPhones, Apple Intelligence is ready instantly — nothing to download. On any iPhone, one tap downloads a small private model that runs fully offline, right from the welcome screen, with progress shown as it lands. Connecting to a model on your computer is still there for power users under Advanced. • Polish throughout Smarter model selection after setup, clearer names for on-device models, and assorted fixes to the first-run experience. As always: private and on-device by default — your data never leaves your phone.

What's New in MyLLM 3.8.1 Your assistant now knows your apps, remembers you, and is steadier everywhere — from the very first message. • A memory that learns MyLLM now quietly picks up the durable things about you from your chats — your preferences, projects, the people and places that matter — and remembers them across conversations, so replies get more personal over time. It all stays on your device, and everything it learns is shown in Settings → Memory, where you can edit or delete anything, or turn auto-learning off. • Your assistant runs MyLLMos with you In chat it now knows your saved mini-apps: ask it to open one ("open my Timer app"), improve one, or hand a document to one through the shared files your apps use. Siri, Shortcuts and the Action Button can launch your apps by name too. • Free, on-device model, front and centre On supported iPhones and iPads, MyLLM works out of the box with Apple Intelligence — no model to download, no API key, fully on-device. This release makes it noticeably smoother, especially on the first message of a chat. • Python and charts you can rely on The on-device Python tool is faster to answer and far more forgiving: better tool-calling with more models, longer time for heavy plots, and clearer errors when something does go wrong. • Your chat history, protected Conversations are now saved more defensively, and MyLLM automatically repairs and recovers its chat index if anything ever goes wrong mid-save — so your history stays yours. As always: private and on-device by default, and a one-time purchase — no subscriptions, no ads.

What's New in MyLLM 3.8 Your assistant now knows your apps, remembers you, and works better with Siri and Shortcuts. • A memory that learns MyLLM now quietly picks up the durable things about you from your chats — your preferences, projects, the people and places that matter — and remembers them across conversations, so replies get more personal over time. It all stays on your device, and everything it learns is shown in Settings → Memory, where you can edit or delete anything, or turn auto-learning off. • Open your apps with Siri, Shortcuts and the Action Button Launch any of your saved MyLLMos mini-apps by name — just say "Open Timer with MyLLM" — or wire one to the Action Button or a Home Screen shortcut. • Siri and Shortcuts that actually run Ask a question, start a new chat, search the web, check email, get your daily brief, summarise a page or start voice input — all straight from Siri, Spotlight, Shortcuts or the Action Button, with your request carried right into the app. • Free, on-device model, front and centre On supported iPhones and iPads, MyLLM now works out of the box with Apple Intelligence — no model to download, no API key, fully on-device. Nothing to set up; just start chatting. • Your assistant understands MyLLMos In chat it now knows your saved apps exist, can open and edit them, and can read and write the shared files your apps use — so it can hand a document to an app, or pick up one an app made. As always: private and on-device by default, and a one-time purchase — no subscriptions, no ads.

Video generation is back — now on Wan 2.2. Our image-and-video server moved to a new, better video model, and this update repoints the app to match so AI video generation works again. • Text-to-video restored — the Video option now runs Wan 2.2, with the right defaults out of the box (1280×704, ~5-second clips). Just type a prompt and generate. • Full control over your own pipeline — Settings → Image & Video Generation now has separate "Custom workflow" slots for image and video, so you can paste your own ComfyUI workflow for either without one affecting the other. • Share your setup — export your server configuration (URL, optional token, and custom workflows) to a single file that another device can import in seconds. Everything still runs on your own server — nothing about your prompts goes through us.

Bug fixes and polish - This is a small maintenance update that fixes two things you told us about: - Smoother tool permissions — when the assistant needs a tool you've switched off, it asks, and the moment you tap Allow it just carries on. No more granting permission only to be told it didn't go through and having to say "try again". - Video playback fixed — YouTube and Vimeo videos now play properly inline in the chat (and in any mini-app that shows one), instead of trying to bounce out to Safari. Thanks for the reports — keep them coming.

Your apps just got more capable — and easier to reach. This update gives the apps in MyLLMos new powers and puts your favourites a tap away. Everything stays on your device, and every capability is yours to switch on or off. • New apps in the gallery — Receipt Logger scans a receipt and tallies your spending, Voice Journal turns what you say into saved entries, and About Me teaches MyLLM your lasting preferences so every chat feels more personal. • Apps can do more — mini-apps can now scan documents, transcribe your voice, and (only if you allow it) add to MyLLM's memory — alongside camera vision and image generation. Each is opt-in under Settings → MyLLMos Apps, and your memories stay write-only to apps: they can add, never read. • Pin your favourites — press and hold any app in MyLLMos and choose Pin. Pinned apps float to the top, and appear as quick-launch shortcuts right on your start screen. • A calmer chat — the apps bar now tucks away while you're typing, so the keyboard never crowds your messages or opens the menu by accident. • Quicker access — new shortcuts on the start screen to open MyLLMos and browse the App Gallery. As always: on-device by default, no account, no subscription. Buy it once, it's yours.

Now MyLLM can see, hear, and remember — privately. This update gives your assistant three new senses and a memory, all working on your device. Nothing here needs an account or the cloud. • Scan documents — point the camera at a page and MyLLM reads the text and adds it to your private knowledge base, ready to search or summarise later. On-device, no cloud. • Ask about what you see — snap a photo and tap a quick question: "What is this?", "Translate", "Read the text". Handy for menus, labels, signs and plants. • Record and transcribe — capture a meeting, lecture or voice note; MyLLM transcribes it on your device and can summarise it with the key points and action items. • A memory you control — MyLLM can remember things about you to personalise its replies across chats. The new Settings → Memory screen lets you see, edit or delete everything it knows, or switch memory off entirely. It never leaves your phone. • Faster image generation — pictures generate quicker, and you can ask for several variations of an image at once to pick your favourite. As always: on-device by default, no account, no subscription. Buy it once, it's yours.

MyLLM now knows where you are — when you want it to. This update gives MyLLM a sense of place. Ask what's nearby, get directions, check the weather, or drop a map into the chat — all on-device with Apple Maps, no account or API key. Location stays private and off until you switch it on. • Find places nearby — "coffee near me", "nearest pharmacy", "Italian restaurants" — with names, addresses, distances and one tap to directions. • Directions & travel times — ask how to get somewhere and get the route, distance and ETA, walking or driving. • Weather — current conditions and a few days ahead, for where you are or anywhere you name. Free and keyless. • Maps in the chat — places and results show on a real map you can open in Apple Maps. • Location-aware mini-apps — three new ones in the App Gallery: Near Me (a local explorer), Essentials Finder (nearest pharmacy, ATM, petrol and more), and Day Out Planner (tell it a vibe, get a route). Any app you build can use your location too — you control it in Settings. • A searchable App Gallery — the gallery's grown, so there's now a search icon up top, plus Update All to refresh your installed apps in one go. • Less scrolling — charts, maps and rendered pages now appear at the end of a reply, right next to where you type, instead of up top. Private by default: location is off until you allow it, never tracked in the background, and you can switch off what apps can use any time in Settings.

What's New in 3.4.1 - Build, fix, and personalise your mini-apps like never before. A REAL CODE EDITOR Editing an app's code now feels like a proper editor: full syntax colouring for HTML, CSS, JavaScript and Python, line numbers, find & replace, code folding, and smart indentation. LIVE PREVIEW See changes as you make them. A live preview pane runs your app right below the editor and reloads as you type — no need to close and reopen. FIX WITH AI Hit a bug? A new debug console surfaces runtime errors, and one tap on "Fix with AI" rewrites your code to fix them — whether or not you write code yourself. A BIGGER, TIDIER ICON PICKER Choose from 150+ symbols plus a brand-new set of colourful app-style icons. Everything is now organised into clear sections with "Show more", so finding the perfect icon is quick and uncluttered. Plus a range of refinements and fixes under the hood. Thanks for using MyLLM. As always, it's a one-time purchase with no subscriptions, no ads, and your data stays on your device.