The Skales blog
Releases explained in plain language, tutorials, honest test reports and videos - straight from the one-person team in Vienna.
Grip is the release where Skales stops when you say stop, and tells you what it costs before it spends. Under every reply sits the price of the session, read from the provider or from the price table, with the sub-agents' bill included; a ceiling you set in Goals stops a run before the next paid step and asks you three things - continue, take a cheaper model, or stop - in the chat, in the Code window, in Iris and on the phone. A typed "stop" is honoured before the first request leaves the machine, and a stopped run says so in its own bubble instead of pretending it finished. Credentials are locked away from the file, shell and search tools; a bulk action or a publish asks first, in every mode, and a yes never covers more than the card names.
Read →Backbone is the release where Skales got one engine. A chat you type, a task you scheduled, a job on a timer, a plugin, a webhook, a Discord message, a Telegram or WhatsApp turn and the desktop Buddy used to run on different loops with different rules; now they all run on the same one, with the same guards, the same budget and the same record of what happened. Close Skales in the middle of a run, update it, let it crash: the run comes back as an unfinished run, says how far it got, and one Continue picks it up at that step. Ask for something big and Skales can hand parts of it to sub-agents, each in its own checkout of your repository, and collect the results. Under every reply sits a context meter that names what filled the window: the system prompt, the tool definitions, your skills, your memory and the conversation itself.
Read →Cockpit 2.0 is named for the screen it replaces. What used to be two stacked tab bars, a second title and a row of icon chips is now one board: everything that stands, in four columns, with goals, your own tasks, Autopilot tasks and schedules as cards you can filter, drag and act on. One head, one action. The phone got the same board in 2.9.21, and a computer on an older version still fills it from what it can tell rather than showing an empty screen.
Read →Slim is named for what it takes away. The sidebar is four entries and a More button you fill yourself, the Teams screen holds one thing instead of two, the Discover composer lost its text box on purpose, and a chat of five hundred turns opens in half the time because the page finally draws only what you are looking at. Underneath that there is a great deal that is new: your agent now carries its working mood in the open, Teams became a real room with attachments and an agent that reads it, ten plugins ship as the point rather than the demonstration, and Skales Code grew a command palette, a project search, hunk-level review and the full git surface. And there is a long tail of repair, most of it in places that had been quietly wrong for a while: the memory only understood English, a file crossing between a phone and a computer was thrown away for a checksum that was never wrong, the local engine took your graphics card and then computed on the processor anyway. So Slim is not a feature release and not a fix release. It is the release where the surface got smaller and the thing behind it got both bigger and more honest.
Read →Everything new in Skales v12.9.15, in plain language.
Read →Until now Skales paired one computer with another. Circle draws a bigger ring: a room holds up to twelve people, every message is encrypted for each member on its own and signed by whoever wrote it, and the agents on your roster can join as members in their own right, running only ever on the machine of the person who owns them. One board per room keeps the work in a single list that nobody starts until everyone has approved it. Around that: plugins arrive as a proper shelf with a gallery, motion videos get a timeline and sound, two teams can finally work at once, and the app knows itself in fewer, truer words.
Read →Every version so far handed you an empty message box and wished you luck. This one does not. Skales opens by asking who you are, a newcomer or someone who has done this before, and reshapes the whole first run around the answer: plain language and a guided provider choice for the one, dense screens and skippable blocks for the other. Theme, accent, city, weather, voice, notifications and safety are settled while you are already there, your phone pairs from inside the run, and the last button plays fifteen seconds of what you just set up instead of dropping you into silence. Docs became Guide, one page that shows what your machine is actually ready to do and puts a real prompt in the composer for you to press send on. Memory gained a search box over everything it holds, and Skales itself gained the same search, so it can find a note it wrote months ago without guessing the name. An idle window is finally idle again, around sixty background timers now stop while you are looking at something else and refresh the instant you come back, and Skales says out loud when it healed itself rather than returning as if nothing had happened.
Read →Skales 12.8.5 "Mirror" makes your agent unshakeable: voice recordings, cancelled uploads and background work can no longer take the server down, and every failure now tells you exactly what went wrong instead of hiding behind a generic error. Your custom agents finally feel like their own: one history across all of them, conversations that answer in their agent's voice, and pictures your agent creates mid-task no longer interrupt the job. Plus rendered videos can go straight to Discover, 3D requests come back as real staged scenes, local speech works on Windows and Apple Silicon, and MCP tools are fully reachable on local models.
Read →Iris is the voice surface in Skales, and since August she has her own place in the sidebar. She listens, she works, she answers - with the same tools the chat has, not a stripped-down subset.
Read →Frontier model quality has largely converged. What decides whether an AI actually gets work done is the harness around it: tools, gates, memory, planning, resumption and surfaces. Here is what that looks like, counted.
Read →Everything new in Skales v12.8.4, in plain language.
Read →Hotfix & Patch Release. We've just released Skales v12.8.3, with a focus on making your daily interactions smoother and more reliable.
Read →Skales 12.8.2 lets your Obsidian vault and your assistant finally live in the same folder: point Skales at the vault itself and every answer reads the notes as they are on disk, writes new ones without ever overwriting yours, and understands "look in my Work vault". The companion's mood is no longer hidden on the Memory page but stands beside you while you work, warm when things go well and cool when they do not, with an honest account of what moved it today. Telegram becomes a real conversation: one message that writes itself as Skales thinks, answers that arrive exactly once, and files that arrive as files. And the app got calmer and quicker where it counts, messages you type while Skales is still thinking queue up and are answered together, opening the app no longer greets you with a wave of pop-ups, and a long draft no longer slows the whole window down.
Read →Two kinds of text get looked at in this release. The first is what a model leaves behind in its own answers: the characters nobody can see and the typography that reads as machine written now come off if you ask for them to, and a single command writes any text again in a plainer, more human voice. The second is what Skales carries about you, which is finally something you can see, shape and delete: your assistant can have a name, a character you set with sliders, and a working history that grows out of the work you actually did together. Around that sits the widest release of the year: real 3D in the chat and in Flow, a video editor, diagrams and coloured code in every answer, and Analyze, a full report on what a conversation cost and where every single token went, worked out entirely on your own machine without calling a model at all.
Read →The sidebar is one list again, the surfaces that had grown into a menu of their own are back where the work happens, and what Skales knows about you can be kept off the network with one switch.
Read →A repair release, built the day after 12.7.4 from what the first devices
Read →This release ships Skales Local, an inference server that lives inside the app. No key, no account, no separate install. Chat, dictation, read-aloud and images can each run on your machine first, with the cloud as fallback, or stay fully local.
Read →A release about seeing what is going on. What is running and where, what a conversation is costing, what a tool actually did or did not do, and what the model wrote when it meant to call one.
Read →A release about being answered. A local provider that would not talk, tools that reported more than they did, and a call you can now interrupt in the middle of a word.
Read →Skales stands for everybody's AI. What it is not yet, it will become. We do not build for a target group - we build for every one of you.
Read →A release about the difference between a screen that says something and a thing that happened. Two answers people were promised, a tool family that reported success it had not earned, and the first accessibility check this app has ever had that actually runs. It also clears out a set of things that were there but could not be got at: an endpoint you had configured and could not pick, a second one you could not find, a number under an answer that read as waste, and a grid you scrolled past every day.
Read →A patch about failures that did not say their name. A browser that would not start, a sign-in that stopped with three words, a settings page that took the whole app down with it, and a picture that came back as an error nobody could read. None of these were new. What was missing in every case was a sentence.
Read →The release after the big one is where your first week of feedback lives. The provider page you can finally read, agents that take turns on one graphics card, a scheduler that owes you nothing after a night of sleep, and a dozen small wrongs made right.
Read →Iris Orbit (Beta) arrived early. Planned for v13, shipped today because voice matters to you. Talk to Skales, watch answers take shape, and more: agent teams, MP4 export on your phone. Update to 12.6.0 / 2.6.0 and say hi.
Read →Two corrections to yesterday and this morning. A light window stays light when you have picked your own colours, and connecting a Google Calendar works again on an account made after 2022.
Read →Code stops being a page inside Skales and becomes a program of its own: a window built for working on a repository, in the shape of a coding tool rather than a chat. Point it at a folder, watch it read, plan and ship the diff, review every change beside the transcript, and tell it once how you like to work. It runs the same engine as everything else and none of the identity.
Read →Being asked a question is no use if the form has nowhere to put your answer. Skales could ask you four things and accept none of them, and the setup screen for browsing the web reported itself ready on the same page that said it was not. Both were the same kind of fault: something reported a state instead of checking it.
Read →Flow got a lot bigger in 12.5.5. 🦎 A deck now leaves as PowerPoint with real text boxes your client can retype, a document as Word with headings and tables intact, and a course as SCORM for Moodle and other LMS. Or take the whole project as one ZIP and hand it to someone else. Motion learned techniques, not just palettes. Fourteen of them, papercut, pixel, particles, whiteboard, kinetic type, sketch, ASCII, isometric, blueprint, neon, comic, origami, silhouette and motion graphics, plus stop motion, glitch and film grain stacked on any of them. Each brings the drawing it needs, so pixel art is really drawn on a pixel grid. And every finished render is now read back frame by frame, so a clip that freezes or a stretch that comes out black is caught in the file, with the exact second it happens.
Read →Sending a picture is a normal chat turn now. It was not: it ran on a path of its own, outside everything the app does to keep an answer alive, and the previous release patched that path instead of removing it. The path is gone, and the same class of loss is fixed everywhere else it hid - in Buddy, in the AI Pointer overlay, and in the model recognition that quietly shrank a new model's context window to an eighth of its real size. Moving that turn onto the normal path also uncovered a promise nobody could see from the outside: an incognito chat with a picture in it was being written to disk. An incognito chat now runs on a conversation that has no file at all. Moonshot AI (Kimi) joins as a provider.
Read →The 12.5.2 reasoning guard cut healthy long-thinking runs. It is behavioural now, the stall card does what it offers, and image turns stop ending in silence.
Read →Connected MCP servers are visible to the AI again, background tasks stop running in circles, slow models get the time they need, and the Tasks page becomes something you can actually keep tidy.
Read →It ships with a library of working skills it can reach for on its own. Some questions now cost you nothing at all. Ask "what's today's date", "what is 3 + 3", "how many kilometers is 12 miles" and the answer comes back instantly, straight from the app, without a single token going to a model. The reply wears a badge so you can see it was free.
Read →Remote access is one switch, default off, and token-protected the moment you turn it on, and pairing now asks you before any device connects.
Read →Skales 12.4.0 is here. Background and multi-agent runs stop paying for the whole tool catalog on every step, the biggest hidden cost gone. Media cards can no longer be faked, and publishing now does SFTP with host-key pinning. Restart Skales to update.
Read →Everything new in Skales v12.3.5, in plain language.
Read →Flow checks its own work before handing it to you, long conversations keep their task, the calendar tells you what is actually wrong, and Skales stops getting in its own way mid-work.
Read →Your phone can now start, watch and stop the work your desktop does. Agents know who you are and what day it is. Vision lands on the right provider, the composer never stays stuck, and Flow stays out of your chats.
Read →This update focuses on making your interactions smoother and more reliable, especially when working with images and different AI models.
Read →**Freeze** is a steady release: tools that get the time they need, backups that survive a move to a new computer, and a round of fixes across Ollama, OpenRouter and Flow. It ships alongside a new mobile release too.
Read →Flow is here. Tell Skales "a 5-slide pitch deck for a coffee subscription, warm colors, bold numbers" and watch it design real slides with a live preview, then keep editing by talking. Eight modes, from decks to motion graphics that render to MP4. Find it in Studio, works with the models you already use.
Read →Per-chat reasoning-effort control (native output_config.effort and
Read →Replies no longer freeze, reasoning models get the time they need to finish, and long tasks that write lots of files now run all the way through. Your reminders, goals and memory are far more dependable too, so nothing silently gets skipped or scrambled. And other improvements.
Read →Skales Visuals grows into a real image and video studio, big files finish instead of stalling, images you send hold up across chat and channels, and the whole run is steadier under a flaky network.
Read →Two people, one plan, each in their own thread. Plus your ChatGPT subscription doing real work, document reading, and a faster start.
Read →v11.4.50 is the steady one. A long scheduled task no longer risks running twice, an unexpected restart keeps your usage limits and recent activity intact, and the Telegram status finally tells the truth about whether your bot is reachable. A calmer, more dependable Skales.
Read →Reliability fixes, Gemini back in chat, correct image ratios in Studio, Teams improvement. Grab it via auto-update or skales.app.
Read →Connect any REST API from its docs with the new API Connector, generate images and video right from the chat composer (now with fal.ai), and image requests work again across chat, WhatsApp, and Telegram. Update from the menu or grab it at skales.app.
Read →Your agents used to stop after about ten steps and call it done. Not anymore. Scheduled tasks, autopilot, and anything you send from Telegram, WhatsApp, or the Buddy now run real multi-step work all the way to the end, as deep as a goal.
Read →You can now paste an image directly into the starting chat box, just like you could in an ongoing conversation. This makes it even quicker to share visual information with Skales.
Read →Your agent finishes the job now: set a Goal step budget in Settings → Goals (or 0 to let a goal run straight to the end while you are away), and it picks up where it left off instead of starting over. New AI models tune themselves automatically, so a fresh release works the day it ships. Plus Veo 3.1 and Nano Banana image models in Studio and chat, a more reliable Google (Gemini) connection, and Wrapped badges that match a big week.
Read →Our free built-in trial model, is rolling out as an early Alpha. Open Skales and start chatting with no API key. It runs on our servers with limited capacity, so it may be paused or adjusted while we tune it. You can change any time to your default provider (BYOK or Local).
Read →v11.4.10 makes scheduled reminders and tasks fire reliably on any model (and on a backgrounded Mac), wires up the integrations that were only promised, and adds Studio Type, a Friendly/Business tone for Friend Mode, and a My Posts view in Discover.
Read →Turn a line of text into an animated, looping video right inside Studio, plus more reliable tool use on any model and a batch of long-standing rough edges gone.
Read →Chat now codes like a proper editor: inline colored diffs, a project map so the agent finds the right file, and a Chat/Code/Plan/Auto switch right on the start screen. Reminders got dependable too, a missed one catches up the next time you open Skales.
Read →This release is mostly about the things you don't see — the work Skales does for you in the background. It now runs reliably, every launch. And your private Briefing no longer hides in its own corner.
Read →Two things in one: skales.app got a full redesign - a real website for what Skales has become - and under the hood, the agent itself got a stability pass. Less magic talk, more things that just work.
Read →One place to manage notification types. The same per-type switches lived twice: on the Notifications page and again under Settings as "Buddy Intelligence" checkboxes - both wrote to the same setting.
Read →Skales shipped 40+ releases while its website stayed a one-pager. Here is the new multi-page skales.app - honest comparisons, a models hub, this blog, and a Buddy that answers questions right on the site.
Read →
The cursor-anchored quick-ask overlay exists twice - as the free standalone app at aipointer.app and built into Skales. Same idea, different superpowers. A practical guide.
Read →The most-requested feature of the last twenty updates - generate your own animated pixel buddy right inside Skales, with live preview. Skales, Bubbles and Capy stay the stars; now they have a wardrobe.
Read →Tool-calling reliability varies wildly between models. LLM Profiles tune tool budget, prompts and sampling per model family - so even a small local model stops fumbling. Here is how it works.
Read →No paid plan, no GPU farm, no trial countdown. Three honest ways to run a full AI agent at zero cost - fully local with Ollama, free cloud tiers, or the ChatGPT subscription you already have.
Read →We use one anonymous analytics cookie to understand what helps visitors. Your Skales app never phones home either way.
