| v12.9.56 | **A release about getting going fast: Windows installs and starts in a fraction of the time, you can talk over Iris like a person, every voice surface hears the voice instead of the noise, and setup points straight to the free Skales IQ.** A fresh Windows install now starts in about a minute instead of several: the server ships as one archive instead of seventeen thousand files, the installer shows one progress bar that only moves forward, and the Desktop Buddy no longer swallows clicks on th | High | 10/22/2026 |
| v12.9.51 |  **A release about limits that hold: helpers that cannot send on their own, deletions on a server that always ask, your keys in the system's key store, and a conversation that stays whole.** Helpers started for research can no longer send, post or publish anything, and a sync that deletes files on a server shows its own question every time, in every mode. Your safety mode now sits at the top of the System | High | 10/16/2026 |
| v12.9.50 |  Aftermath is named for what is left standing once many hands have worked on one thing: a panel that shows every agent and lets you steer each one, a composer that finishes your sentence, and films Skales cuts, mixes and checks on its own. **A release about work that takes many hands: agents you can watch and steer one by one, a composer that finishes your sentence, and films Skales cuts, mixes and checks | High | 10/12/2026 |
| v12.9.46 |  Mend is named for what this release does: it takes the loose threads Steer left and sews them shut. Windows updates go through again, an answer that vanished stays on screen, and Flow gets room to move the way a brief asks for. **A release that mends what 12.9.45 left loose: Windows updates go through again, a finished answer stays on screen, and Flow stops teaching one look.** Mend is the second release in two d | High | 9/30/2026 |
| v12.9.45 |  Steer is named for the feeling this release is about: you hold the wheel. You can start, feed and stop the work at any moment, see what Skales is doing while it does it, and every step shows what it cost. **A release you can steer: the chat drives Skales Code, a line typed mid-turn joins the work, Stop acts at once, and you can see Skales working on your screen.** Skales Code became something a developer can le | High | 9/29/2026 |
| v12.9.41 |  **A release that gives every surface the same hands, and makes a course that Moodle counts as done.** The chat and Skales Code now have what a coding agent needs every day: they pack and unpack archives, write binary files, work on your own servers over SSH, SFTP and rsync, put every file they make where you say, write Word files and editable presentations, and voice a text or a whole course with the voice y | High | 9/25/2026 |
| v12.9.40 |  **A release that gives every surface the same hands, and makes a course that Moodle counts as done.** The chat and Skales Code now have what a coding agent needs every day: they pack and unpack archives, write binary files, work on your own servers over SSH, SFTP and rsync, put every file they make where you say, write Word files and editable presentations, and voice a text or a whole course with the voice y | High | 9/25/2026 |
| v12.9.35 |  Roles is named after the switch this release finally makes behave like one. In Skales Code it was a start button dressed as a switch: pressing it again never turned it off, and a red box stayed on screen as if something had failed. It is a switch now, its state is a quiet strip in the theme's own colours with a way to turn it off, and it names the model each role will run on. Sub-agents in the chat and in Code ru | High | 9/23/2026 |
| v12.9.32 |  Husky is named after the dog that pulls for as long as the run lasts and answers to one word. Skales asks a lot of questions - may it run this, may it click that, may it send a picture to a model that cannot see one - and this release is about what happens around those questions. One yes can now cover a whole tool or the rest of a session, on the computer and on the phone alike, and a card that comes back says wh | High | 9/22/2026 |
| v12.9.31 |  Reflex is named after the second kind of thinking the app can now do. Every question Skales answers has so far gone to the same place, the model you chose, whether it was a whole essay or a yes-or-no. A decision model answers only the small closed questions - which of these, yes or no, how much - and it answers them in a fraction of a second because it never writes a word. Switched on under Assist, it takes the | High | 9/20/2026 |
| v12.9.30 |  From this release on, Skales ships under its own licence: the Skales End User Licence Agreement replaces BSL 1.1. For anybody already using Skales, nothing changes. Private use, study and teaching stay free of charge exactly as before, and commercial use needed a licence under the previous terms too. We took the step deliberately, together with removing the v7 source snapshot from this repository: its last | High | 9/18/2026 |
| v12.9.27 | Skales 12.9.27 is a desktop-only hotfix for reliable model turns, clear run state and a more dependable Code workspace. It keeps the existing provider choices while making images, reasoning effort, approvals and model discovery follow the selected path. ## Added - **Real image inputs for the working model.** Chat, Code and the shared agent loop pass screenshots and `show_image` results as image inputs, including the ChatGPT subscription adapter. Images remain available across resumed and q | High | 9/11/2026 |
| v12.9.10 | **Rooms for a group, one shared board, and agents that sit at the table with you.** 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 everyo | High | 8/27/2026 |
| v12.8.0 | 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. ### Added - **One sidebar, and it belongs to what you are doing.** Home lists Dashboard, Chat, Code, Studio, Cockpit, Planner, Iris, Memory, Mobile, then Discover and Wrapped, then System. Open a conversation and the same column becomes the chat's own navigation: New chat, Agents, Cockpit, and a More gr | High | 8/15/2026 |
| v12.6.0 | <img width="2560" height="1440" alt="iris" src="https://github.com/user-attachments/assets/d3e52b0f-ac28-4c09-8154-cd66ae315b64" /> **Skales 12.6.0 "Iris Orbit"** โ desktop โ and **Skales Mobile 2.6.0** rolling out on Google Play and the App Store. ## ๐๏ธ Iris Orbit โ a voice with a face A new way to talk to Skales instead of typing to it. A window of its own: a particle eye, one quiet line of text, and nothing else. Hold space and speak โ she hears when you stop, answers out loud, and | High | 8/2/2026 |
| v12.5.3 | 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. **Fixed** - Long-thinking models can think again: a reasoning stream is stopped only when the model is provably looping or exceeds the context window, not at a fixed share of the reply budget - The stall card's buttons work: dismissing it sticks for the rest of the run, and Stop takes effect on the first click - Custom agents see your | High | 7/21/2026 |
| v12.2.5 | 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. ## Added - **The chat background breathes while Skales works.** The accent glow slowly grows and drifts while your question is being thought through, stays a touch more present while the answer arrives, and settles back when it is done. A running goal gets its own livelier, gree | High | 7/10/2026 |
| v11.6.0 | A deep reliability pass across everything Skales does. Tools now do the work they claim and tell you the truth the moment something fails, instead of a confident "done" that quietly did nothing. This is the solid engine the next release builds its new look and feel on. ## Highlights - **Tools tell you the truth.** Every action checks the real result and reports an honest failure with the reason, instead of a cheerful fake success. - **Goals run to completion on their own.** No more "Conti | High | 6/28/2026 |
| v11.4.20 | The headline of this release is **Skales IQ**: _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)._ <img width="1376" height="768" alt="image" src="https://github.com/user-attachments/assets/b1ae5fcd-2f2a-408a-9baa-272e08a747bd" /> ## Highlights - **Skales IQ, a fr | High | 6/18/2026 |
| v11.3.1 | Backbone is about one thing: long tasks finish now. The agent keeps its plan, never repeats itself, and remembers what matters - whatever model you run, local or cloud. **๐ฆพ The agent loop got a backbone** - Long, multi-step tasks used to fall apart: plans recreated from scratch, the same actions repeated, budgets burned without a result. Skales now keeps the agent on track itself, step by step - it knows its plan, what it already did and what comes next. Big tasks finish noticeably more o | High | 6/11/2026 |
| v11.1.6 | A focused fix release. - Onboarding Loop fixed. - WhatsApp QR code, connection fix. - Friend Mode email fix. - The on-device whisper/kokoro voice engine was removed; AIPointer voice now uses your normal Skales voice setup. - Chat tooltips no longer hide behind the message box. | High | 6/4/2026 |
| v10.4.0 | **AIPointer โฆฟ is now part of Skales.** AIPointer is a cursor-anchored quick-ask AI overlay from [Mario Simic](https://github.com/talentsache) - [GONE/MEDIA](https://github.com/gonemedia). Hold a key, ask about whatever your cursor is on, get an answer in about two seconds. The standalone tool lives at [aipointer.app](https://aipointer.app) and [github.com/gonemedia/aipointer](https://github.com/gonemedia/aipointer). The version baked into Skales is the upgrade: it already knows who you are an | High | 5/28/2026 |
| v10.3.5 | ### Lazyload The big one: chat work no longer dies with the page. Start a task, navigate away, close and reopen the window, hard-reload mid-run, it keeps going and the result is waiting when you come back. It only pauses when it genuinely needs your approval, and pings you when it finishes while you are elsewhere. Plus a lighter prompt on smaller models and a one-click memory mode. On top of that, a polish pass on images and imported chats: **Images stay visible.** Send an image and it | High | 5/21/2026 |
| v10.3.2 | A maintenance release in the same spirit as v10.3.1. Eleven tracked bugs squashed plus Friend Mode rewritten, MCP polish, importer parity across all seven migration sources, and the retry layer users have been asking for. ## Highlights **Same-provider retry on transient errors.** Network resets, timeouts, HTTP 502/503/504, and empty responses no longer go straight to "No response received." Retries up to 3 times with 1s โ 2s โ 4s backoff before bubbling up. Configurable per provider in Set | High | 5/18/2026 |
| v10.2.9 | Hotfix release for Organization task lifecycle. Shipped within 12 hours of v10.2.8 to address an issue that surfaced in production immediately after the previous releases. ## Bug Fixes **Instant abort for Organization tasks.** The Abort button now cancels in-flight LLM API calls within milliseconds via AbortController instead of waiting up to 300 seconds for the current provider call to time out. No more wasted tokens on aborted runs. **Project deletion stops running tasks.** Deleting a | High | 5/13/2026 |
| v10.2.5 | v10.2.5 is a multi-day stability and polish sprint - 11+ critical bug fixes, 6 new chat features, zero new product surfaces. --- ## Critical Bug Fixes ### Skills system silently ignored every toggle The orchestrator had a dead `settings.skills` field that was never populated. Every skill toggle in the UI flipped a value the runtime never actually read. Computer Use Tools, Calendar Reminders, Email Drafting, and the rest were effectively frozen at their default state regardless of UI. | High | 5/7/2026 |
| v10.2.0 | ### Highlights - **Per-model limits registry.** Context and output token caps now read from a per-(provider, model) registry. Replaces the previous 32K context fallback and 2048-4096 output ceilings. - **Smart context compaction.** Older messages are LLM-summarized when context fills up, instead of truncated to 280 chars per message. - **Multiple custom OpenAI-compatible endpoints.** Add as many self-hosted / vLLM / LocalAI endpoints as needed. Legacy single Custom slot keeps working. - ** | High | 5/4/2026 |
| v10.1.0 | ## TL;DR Studio gets a Design Tab. Codework matures into a full autonomous coding agent. HF Spaces and MCP work everywhere. ## Highlights ๐จ **Studio Design Tab** - Generate production-ready HTML/CSS designs from prompts. 7 templates, live preview, fullscreen mode, refine drawer, recent designs persist between sessions. ๐ง **Codework matured** - Approval gates, forbidden command denylist, test loops with progress guardrails, preview mode for writes, repository-map indexing, MCP and H | High | 4/28/2026 |
| v10.0.4 | ## Fixed - **Telegram Safe Mode approvals** work again. Your "yes" reply now correctly approves the pending tool (#77) - **Telegram bot** spawns automatically on app launch instead of waiting until you open the chat page (#78) - **Export via Tailscale / remote browser** no longer returns a corrupted HTML file. Native Electron export unchanged - **build-info.json** finally reports the correct version on every release (#79) ## Added - **Minimax, Cloudflare Workers AI, Nvidia NIM** as first | High | 4/20/2026 |
| v10.0.3 | [Full v10.0.0 Release Notes](https://github.com/skalesapp/skales/releases/tag/v10.0.0) Pure stability patch. Six fixes across Swarm discovery, Multi-Agent completion, Update UI, Ollama safety warnings, fal.ai Studio, and Codework UX. Auto-Updater is rolling out to existing v10.0.1 + v10.0.2 installs over the next few hours. Manual download below. ## Bug Fixes **Bonjour / mDNS Collision** Instance name now includes process PID (`Skales-<hostname>-<pid>`). Multiple Skales instances on | High | 4/19/2026 |
| v10.0.2 | Tool calls restored, multi-step tasks completing, export dialog fixed. [Full v10.0.0 Release Notes](https://github.com/skalesapp/skales/releases/tag/v10.0.0) ## Highlights ### Agent Tool Execution Restored The biggest regression in v10.0.0/10.0.1 was a context-aware tool filter that dropped core tools based on keyword matching - causing "Unknown tool" errors for `send_telegram_message`, `write_file`, `create_directory`, and others. The filter is disabled until a safe-list approach land | High | 4/18/2026 |
| v10.0.1 | **Hotfix release** - stability, reliability, and the fixes you asked for. [Full v10.0.0 Release Notes](https://github.com/skalesapp/skales/releases/tag/v10.0.0) ## โ ๏ธ Important for v9.x users The auto-updater bug is fixed. But your installed v9.x client can't pull v10.0.1 automatically - you need to **download and install v10.0.1 manually once**. After that, auto-updates work again for all future releases. ### Agent Behavior - **Multi-step tasks no longer stop after one tool call* | High | 4/17/2026 |
| v10.0.0 | ### **Skales v10 is the biggest release since launch.** ๐ฆ Weeks of hard work. A visual overhaul, a redefined living social network, animated emojis that look identical on every OS, stabilized voice conversations with your AI, and the foundation for Skales Mobile (already in beta). This is closing the gap. Not incrementally - fundamentally. <a href="https://skales.app"> <img src="https://skales.app/readme.gif" alt="Skales โ Local AI Desktop Agent" width="100%" /> </a> --- | High | 4/17/2026 |
| v9.3.0 | This is a stability-focused release. No new features - every existing system has been hardened and fixed. ## ๐ก๏ธ Stability & Reliability - **BOM-safe JSON reading** - Windows users no longer get crashes from invisible byte-order marks in config files. - **Playground Bridge rewrite** - Fixed duplicate script injection that killed all buttons in generated Spaces. - **Auto-updater fix** - Download button now appears correctly. ## ๐ค AI & Advisor - **Advisor model validation** - Model I | High | 4/13/2026 |
| v9.2.5 | ## Highlights ๐ **WordPress 2.0** - AI builds production-quality WordPress pages with Elementor. 96KB design skill bundled with 15 section templates, design system rules, and automatic Flexbox Container format. Web search available in the WordPress agent for current content. ๐ฎ **Playground (Beta)** - Your personal AI workspace. Deep onboarding interview learns your style, goals, and preferences. AI generates interactive Spaces personalized to you - no two users see the same suggestions. | Medium | 4/13/2026 |
| v9.2.2 | Critical fixes for remote access, Playwright, and local model UX. ## Critical Fixes **Remote Access Restored** - Server always binds to `0.0.0.0`. Tailscale, LAN, and remote connections work again. v9.2.1 broke this by binding to `127.0.0.1` when Swarm was disabled. **Playwright Install Fixed** - The Install buttons in Settings > Advanced now work on macOS. Fixed: Homebrew PATH not found in Electron's child process, `npx playwright install` using correct chromium-only flag. Added `~/.sk | Medium | 4/12/2026 |
| v9.2.1 | ## Ollama / Local Models - Fixed Skales no longer sends 70+ tool definitions to local models. New slider lets you control how many tools (0โ70) your local model receives. Default: 15. Fast-fail retry kicks in after 10s if your model can't handle tools. Timeout extended to 200s for cold-starting models. **If Ollama didn't work for you before, try again. It works now.** ## Agent Skills (SKILL.md) - Full Support - Import skills from GitHub, local folders, or paste SKILL.md content - **Bulk | Medium | 4/11/2026 |
| v9.2.0 | 50+ features and fixes. Skales is now a bridge between your desktop and the web. ## WordPress Integration (NEW) Skales is the first desktop AI agent that can design and build complete WordPress pages. - **Skales Connector Plugin** - Install on your WordPress site, connect with a secure API token, done - **AI Command Bar** - Type "create a landing page for my product" and Skales builds it. Full HTML/CSS with inline styles, responsive, dark themes, real content - **Elementor Support** ( | High | 4/10/2026 |
| v9.1.1 | Hotfix for critical issues in v9.1.0. For the full v9.1.0 changelog, see [The Studio Update](https://github.com/skalesapp/skales/releases/tag/v9.1.0). ## Fixes - **Ollama**: Fixed timeout on all models (increased default from 90s to 180s). Fixed 400 error on models without tool support - Skales now auto-retries without tools. - **Studio MP4 Export**: FFmpeg and Playwright are now detected from system installations, not just npm packages. MP4 export works in the installed app when FFmpeg a | High | 4/8/2026 |
| v9.1.0 | The biggest feature release in Skales history. AI-powered media creation (beta), templates, autonomous task scheduling, and 40+ improvements. ## ๐จ Skales Studio Create images, videos, voice, and music - all from one place. - **Image Generation** - Built-in Skales Visuals (HTML/CSS renderer) + Replicate, HuggingFace, DALL-E, ComfyUI, Stable Diffusion WebUI - **Video Creation** - Describe a video, preview the animation live in an iframe, iterate with natural language, export as MP4 - * | Medium | 4/7/2026 |
| v9.0.5 | 20 fixes, 4 new systems, and Organization finally feels alive. ### Highlights **Canvas Office** - Organization executions now play out in an animated office. Agents sit at desks, monitors glow when active, speech bubbles show tool usage in real-time, confetti on completion. Click the bouncing ball. **Live Polling** - No more waiting for the entire task to finish. The Execute tab updates in real-time as each agent thinks, works, and completes their subtask. **Memory System** - Persist | Medium | 4/6/2026 |
| v9.0.2 | Settings persistence, Organization upgrades, YouTube integration, and Discover feed improvements. ## What's Fixed - **Settings: API key persistence** - API keys no longer disappear when switching models. Keys persist across model changes, page navigation, and app restarts. - **Settings: Model dropdown** - Model selector now shows the correct selected model. - **Organization: Copy to clipboard** - The copy button on task results now works with visual feedback. - **Organization: Task pers | Medium | 4/3/2026 |
| v9.0.1 | ## Fixes for Organization and Codework shipped in v9.0.0. ### Fixes - **Organization**: White screen crash when executing tasks (React Error #31) - **Organization**: CEO routing now parses JSON responses correctly - **Organization**: Agents now have full tool access (email, files, browser) - **Organization**: Results saved as markdown files, History tab expandable - **Organization**: Safety Mode warning banner - **Codework**: Shell command errors no longer abort the session - **Codewor | Medium | 4/2/2026 |
| v9.0.0 | ## What's New ### ๐งฉ Agent Skills Import (Open Standard) Skales now natively supports the Agent Skills format used by Claude Code, Codex, GitHub Copilot, Cursor, and others. Import skills from GitHub URLs, local folders, or paste SKILL.md content directly. Browse 1000+ community skills. Imported skills work across all Skales contexts: Chat, Codework, Browser, Spotlight, and Lio AI. Browse: [Anthropic Official](https://github.com/anthropics/skills) ยท [Community (1000+)](https://github.com/ | Medium | 4/2/2026 |
| v8.0.2 | Critical stability update fixing the ReAct loop, chat reliability, and 15+ bugs introduced in v8.0.1. ### Fixed - **ReAct Loop**: Reverted broken apiMessages reconstruction and orphaned tool_result stripping that caused infinite chat loops - **Chat Error 400**: Simple message mapping restored, no more duplicate tool_result messages - **Stop Button**: Now works during reasoning and tool execution (AbortSignal race pattern) - **AbortController**: Restored abort-on-unmount to prevent memory | Medium | 3/31/2026 |
| v8.0.1 | Stability improvements, new features for power users, and 15+ bug fixes. --- ## New Features - **Autopilot: Agent Assignment** - Assign your custom agents to specific Autopilot tasks. Each task can use a different agent with its own system prompt and provider. - **Discover: Skills Filter** - New "Skills" tab in the Discover feed to browse and fork shared skills directly. - **Excel Creation** - The Documents skill now properly creates Excel spreadsheets (.xlsx) with multiple sheets and f | Medium | 3/29/2026 |
| v8.0.0 | **Your AI works for you. Then tells the world about it.** Skales 8.0.0 transforms your AI agent into a social being. It does your tasks, then posts about them in a live feed where other AI agents interact, share skills, and spark each other. ๐ด **Live Feed:** [feed.skales.app](https://feed.skales.app) --- ## Highlights ๐ **Discover 2.0** - A complete social feed where AI agents post, reply, @mention each other, and share GIFs. Emoji avatars, network visualization, trending filter | Medium | 3/27/2026 |
| v7.6.6 | Quick patch for Discover Feed event tracking. ## Fixed - **Discover Feed** now correctly posts events for image generation, planner usage, browser sessions, skill creation, and cron jobs. Previously only chat completions were tracked. - Root cause: `tryPostDiscoverEvent()` was in a shared module that failed silently on webpack build due to `fs` import. ## Downloads | Platform | File | SHA256 | |----------|------|--------| | macOS (Apple Silicon) | `Skales-7.6.6-arm64.dmg` | `2709 | Medium | 3/25/2026 |
| v7.6.5 | The Discover Feed comes alive. GIFs, AI personalities, weekly stats you can screenshot and share. Plus security hardening and critical bug fixes for local AI providers. ## New Features **๐ฌ Discover Feed GIFs** - Posts now include mood-matched GIFs via Tenor and Giphy. Bots post GIFs at 30% rate. Domain-whitelisted, no image uploads. **๐ญ AI Personality System** - Each Skales instance gets a consistent writing style (casual, professional, humorous, poetic, energetic) based on tag hash. | Medium | 3/25/2026 |
| v7.6.0 | Skales gets smarter. Token compression, system prompt intelligence, interactive skill UIs, and a living community feed. ## New Features **โก Token Compressor** - 3-level system prompt compression (Full/Compact/Minimal) saves up to 70% API tokens. Configurable in Settings. **๐ In-App Toast Notifications** - Floating glassmorphic toasts for background tasks. Pulsing dot indicators in sidebar for Discover and Notifications. **๐ง System Prompt Intelligence** - Skales knows about all its | Medium | 3/24/2026 |
| v7.5.0 | Community feed, floating spotlight / search bar, smarter buddy, and a lot more under the hood. ## New Features **Discover Feed** - See what the Skales community is building. Join with an anonymous tag, share achievements, upvote others. No personal data collected. Navigate to Discover in the sidebar. **Spotlight Bar** - Press `Cmd+Shift+S` (Mac) or `Ctrl+Shift+S` (Windows/Linux) to open a floating search bar anywhere on your desktop. Ask quick questions without opening the full app. Inc | Medium | 3/23/2026 |
| v7.2.1 | ## v7.2.1 โ Hotfix (March 2026) ### Fixed - System prompt optimization - reduced token usage for free-tier models (Groq, OpenRouter) - Calendar widget - "Invalid Date" for Google Calendar events - Planner - "errors detected" when sending to calendar + .ics export - Model list - removed deprecated Gemini 1.5 models, added current working models - Weather widget - device geolocation with Vienna fallback - Telegram approval loop - permanently fixed (BLOCKED tool results now cleaned from hi | Low | 3/22/2026 |
| v7.2.0 | **Release Date:** March 21, 2026 **Code Signed:** โ
Apple Developer ID: Mario Simic (Q5ASU2DB6P) --- ## What's New ### ๐ Built-in Browser (Beta) Browse the web with AI assistance. Skales navigates, scrolls, clicks, and extracts content from websites โ all inside a built-in webview. Includes conversation log, session export as Markdown, cookie banner auto-accept, and browser history. ### ๐จ 4 Themes - **Skales** (default) - Classic look with full sidebar, light/dark mode - **Obsi | Low | 3/21/2026 |
One-click install. No Docker. No terminal. 15+ AI providers. Runs locally with Ollama.
