What's new

Every release, newest first. Ocadox updates itself: you get each one from inside the app. Download the latest.

0.10.11 October 21, 2026

GPU: one job at a time on the graphics card, first come first served.
Music: songs sung in your own voice.
Tests: check a model against every agent tool, on GitHub.
Brand: Oda without feet.

0.10.10 October 20, 2026

Video: choose LTX-2.3 (with sound), HunyuanVideo 1.5 or LingBot-Video next to Wan 2.2.
Video: recommend a video model for your graphics card.
Video: long videos for YouTube, 1 to 30 minutes, narrated in your voice.
Brand: Oda icons for Android and iOS.
Tools: start on a model on this PC, not the cloud chat model.
Tools: start on the biggest model on this PC.
Claude Code and Codex: pass on a tool's instructions.

0.10.9 October 19, 2026

Tools: the menu's text is white in the dark theme.
Tools: a Terminal you type in.
Agent: read and type into the Terminal tool's tabs.
Install Ollama from the welcome screen on every system.
Knowledge: chats that answer from your own documents, with sources.
Compare: several models answer one prompt side by side, with what each cost.
Tests: give each Notes test its own folder.
Share your models: an OpenAI-compatible API, paced by the fan guard.
Phone page: work on iPhone, over plain HTTPS when WebSockets are refused.
Quick Ask: a hotkey in any app opens a small box with the selected text.
Research: a question in, a cited report out, read on this PC.
Audio overview: two hosts talk your documents through, like a podcast.
Data analysis: ask a spreadsheet questions, answered with tables and charts.
Translate a document, keeping its layout where the format allows.
Notes: open an Obsidian vault and keep its pages as Obsidian wrote them.
Brand: put the O logo and Oda in the app.
Brand: put the O logo and Oda in the phone app.
Brand: the O logo and Oda on the website, Oda in Notes and Bots.
Brand: build the app icons straight from docs/brand/icon.svg.
Agent: accept common argument names and explain missing ones.
Brand: Oda is the app icon.
Brand: Oda in the top bar instead of the O.
Build: rebuild the exe icon when the icons change.
Bots: no Oda in the empty list.
Brand: the wordmark in the top bar.
Tools: a /command for each one in the chat.
Quick Ask: open from the Tools menu.

0.10.8 October 18, 2026

Speech: clone from a 12-second clip of the voice sample.
Agent: local models keep going and compact their steps when memory fills.
Video: one at a time, and upright photos animate upright.
Tools: the search box keeps what you type.

0.10.7 October 18, 2026

Chat: type your next message while a reply is running and press Enter; it's queued and sent when the reply finishes.
Models: back to just Opus, Sonnet, Haiku and Fable when signed in with Claude Code.

0.10.6 October 18, 2026

Models: every current Claude model is listed when signed in with Claude Code.
Terminal: ocadox models also lists the image, video, music and speech models that are set up.

0.10.5 October 18, 2026

Video: /video_create starts from an attached picture, and attached photos have Animate.
Models: Fable 5.1 when signed in with Claude Code.
Terminal: ocadox run works with every model from the app; ocadox models lists them.

0.10.4 October 18, 2026

Settings: choose where image and voice models are stored, like another drive; existing models move there.
Images: /image_create uses attached pictures as references.
Hosting: keeps listening after a failed connection and restarts cleanly on the same port.

0.10.3 October 18, 2026

Chats: open, search and start other chats while one is still replying; it keeps going in its own tab.

0.10.2 October 17, 2026

Design: clearer names in Ctrl+K, the tour and help.

0.10.1 October 16, 2026

Notes: clearer names in Ctrl+K and the Tools menu.

0.10.0 October 16, 2026

Notes (Tools > Notes): pages inside pages, a block editor (type / for headings, to-dos, toggles, callouts, quotes, code, tables, pictures and dividers), [[links]] between pages with "Linked from" at the bottom, favourites, search and a trash you can restore from.
Your notes are plain Markdown files in the project's Notes folder: yours to search, back up and commit, with no account and no cloud.
Writing help in any page, with any model on your PC or in the cloud: select text for Improve, Fix, Shorten, Longer, Simplify, Professional, Friendly, Translate, Summarize, Action items, Explain or Continue; or type /ai to have a page drafted.
Databases (Pro): tables of pages with properties (select, tags, date, checkbox, number, person, link), sorting and filtering, and a board view you drag cards across. Ask your notes (Pro) answers questions from everything you've written, with the pages it used.
9 templates: Meeting notes, To-do list and Daily journal are free; Project plan, Task board, Content calendar, Reading list, Weekly review and Brainstorm come with Pro.
The agent and bots can read and search your notes, and (Pro) write them, e.g. a daily log; with "Ask before each edit" on, they ask first. Meeting notes land as pages under Meetings, and any page can go into the chat or to Poster.

0.9.3 October 15, 2026

Design it for me paints the photos too: tick "Paint the photos too" and each photo frame gets the photo the model described, made on your PC and cropped to fit. With Pro, "Use my brand kit" makes the six designs in your brand's colours and font, with your logo in the corner.
Design: pages. Add, duplicate and delete pages at the bottom of the page, for carousels, menus and multi-page flyers; every page follows a change of size.
Design: Download PDF (every page, for printing and sharing) and Download every page (one PNG each).
Design: Download MP4, animated: things come in one after another (rise, fade, pop or slide), every page in turn, for Reels, Shorts and TikTok. Made on your PC with FFmpeg.
Design: photo frames show on light pages too, and an empty frame's "add a picture" words stay out of downloads.
A one-minute tour of the best things in Ocadox after first-time setup; take it again any time from Ctrl+K > Take the tour.
Design it for me no longer makes up dates, places, prices or websites you didn't give it.

0.9.2 October 15, 2026

Design it for me: your model writes the words (headline, subheading, details, button, tag) and picks a mood, and eight designer layouts (split, centred, framed, photo band, colour block, badge, editorial, card) arrange them in that mood's colours, with every line sized to fit. You get six designs; Shuffle shows more with the same words, and changing the size redraws them for the new size.
Design it for me works with any model: one on your PC (with your power cap and governor) or a cloud model (counted in your spending meter). Pick it under "Writes with".

0.9.1 October 15, 2026

Design it for me: describe the design ("a thumbnail for my video about quiet local AI") and your chat's model lays out three for the page's size, with real words, shapes, photo frames and colours. Click one to open it and change anything; tick "Also paint a background picture" for a matching background made on your PC.
Design: 66 templates in categories (YouTube, Instagram, Story, Quote, Sale, Event, Podcast, Banner, Business, Flyer): thumbnails, announcements, tips lists, carousels, polls, Reel covers, coupons, gig posters, invitations, webinars, hiring posts, reviews, menus and more, most in several colourways. 14 are free.
Design: if a picture tool comes back without a picture, you get a plain message instead of an error.

0.9.0 October 15, 2026

Design (Tools > Design): make posts, thumbnails, stories, posters and banners on your PC. Templates, text styles and font pairs, shapes, photo frames, uploads (or drag pictures in), snap guides, zoom, undo, and Share to download a PNG or JPG, save the design to edit later, or put it in the chat.
AI on the canvas: make a background or a picture from a description, remove a photo's background in one click, change a picture by asking, upscale it, and have your model write the words. All local.
Free has four templates and marks its downloads with a small "Made with Ocadox"; Pro has every template, a brand kit (your colours, font and logo) and clean downloads.

0.8.11 October 15, 2026

Presenter videos (Tools > Presenter video, or tick "Put me on a screen recording" in Talking avatar): AI you, in your voice, in a corner bubble over a screen recording, like Loom. Pick a recording or use the one in the chat; choose the corner, size and a circle or square; captions across the bottom. If you talk longer than the recording, it holds the last frame until you finish.

0.8.10 October 14, 2026

Tools: the group headings are easier to read (white in dark mode).

0.8.9 October 14, 2026

Free and Pro: every creative tool still works in the free version, now up to a size: talking avatars up to 15 seconds, songs up to 30 seconds in fast quality, videos up to 10 seconds, dubs of videos up to 1 minute, 2 Shorts at a time, and transcript edits up to 2 minutes. Ocadox Pro (or the 7-day trial) has no limits, and each window says what the free version allows. Anything you've already made is yours.
The monthly spending limit for cloud models is now part of Pro; the spending meter stays free.
A new Tools launcher: the tools grouped into Create, Edit, Record and Work, with a search box (type and press Enter), your last three tools on top, and a tag on any that need setting up first.

0.8.8 October 14, 2026

Your own Anthropic key now uses Anthropic's own API instead of its OpenAI-compatible one: the conversation is cached (repeats billed at a fraction, so long chats cost much less), agent runs keep Claude's reasoning between steps, current Claude models get their full 1M-token window, and Spending shows each reply's cost.
Talking avatar: no more jump every 5 seconds. Each part now starts half a second early and dissolves into the one before over the same words.
Talking avatar: when another program (like Ollama) holds much of the graphics card's memory, a part that doesn't fit is made again within the memory that's free, instead of failing.

0.8.7 October 14, 2026

Talking avatar: fast mode. Add it in Settings > Image creation > Talking avatar: fast mode (1.2 GB) and choose Fast: 4 steps instead of 20, so each 5 seconds takes about 10 minutes instead of nearly an hour on a busy 24 GB card (much less with nothing else using it). The face and lip sync hold up; it's a little less lively.
Claude Code with your Anthropic Console key now counts toward Spending: each run shows its tokens (with what came from the cache) and cost, and the monthly limit applies to it too. A Claude plan isn't counted, since it isn't billed per use.

0.8.6 October 14, 2026

Cloud spending: under each cloud reply, the tokens it used (and how many came from the cache) and what it cost, with the chat's total. Agent runs add up every step.
Settings > Cloud models > Spending: this month's total by model, the months before, and an optional monthly limit. Once it's reached, cloud requests stop (local models keep working).
Prompt caching for Claude models on OpenRouter: the conversation so far is reused from the cache, so each message in a long chat costs much less.
The Memory meter under the message box now uses each model's own window, so a cloud chat no longer looks full at 16,000 characters.

0.8.5 October 14, 2026

Cloud models use their full context window. A chat with an OpenRouter, OpenAI or Anthropic model (and Claude Code or Codex) now carries the whole conversation and is only compacted at about half the model's real window, instead of the small budget meant for your graphics card. Local models keep the quiet budget. /compact still works any time, and keeps a long cloud chat cheaper.

0.8.4 October 14, 2026

Talking avatar (Tools > Talking avatar): a photo of you says a script in your saved voice, or a recording, with the lips in time, like HeyGen but on your PC. Vertical or wide, with captions. Set it up once in Settings > Image creation > Talking avatar (about 15 GB; needs a graphics card with 16 GB or more).

0.8.3

Books (Tools > Make a book), for every kind of writing: novels, short stories, poetry, screenplays and stage plays, memoir, biography, history, nonfiction guides, cookbooks, textbooks and workbooks, essays, and picture books. Each is planned, written and laid out its own way.
Pictures and a cover made on this PC, with one of your Characters as the star if you like, and audiobooks read in your saved voice.
History mode, for history, biography and nonfiction: research each chapter on the web (with web access on) or from your own PDFs, Word files and notes; chapters cite their sources; a fact-check flags claims the sources don't back up; and a timeline.
Export a print-ready PDF at Amazon KDP sizes, or an EPUB e-book, and post about it through the Poster. The agent and bots can write chapters and export too.

0.8.2 October 13, 2026

The ocadox command: Settings > Terminal, editor and moving computers > Add ocadox to the terminal, then type ocadox run "task" or ocadox mcp in any new terminal.
Open in your editor: each file the agent (or Claude Code) changes has an Open button that shows it at the changed line in Cursor, VS Code or Zed; Settings picks which.
Move to another computer: save your bots, skills, voices, settings and model list to one file, and bring it in on the new computer (chats, keys and characters if you tick them).

0.8.1 October 13, 2026

Use Claude: an Anthropic Console API key is now the recommended way to use your Claude Code (Settings > Cloud models > Use Claude). Paste it once and Claude Code uses it, billed per use; signing in with a Claude plan is still there as an alternative.

0.8.0 October 13, 2026

Chat tabs: several chats open at once, each in its own folder and running on its own. Press + or Ctrl+T for a new tab. A dot shows a tab that's working, finished, or waiting for your OK; closing a tab keeps the chat in Chats.
Long jobs (video edits, Shorts, music, dubbing, voice-overs) put their result in the tab they started in, even if you've moved to another.
Local models take turns on the graphics card, so the fan guard keeps working; cloud models (Claude Code, Codex, OpenRouter) run side by side.

0.7.9 October 13, 2026

Setup picks extras for your PC: a new first-run step recommends what fits your graphics card (voice, speech, images, music, picture editing, video), downloads it in one go, and offers Make a picture and Make music as first tries. Run it again any time from the command palette.
Use Ocadox from Claude Code, Cursor or Codex: `ocadox mcp` gives other agents Ocadox's local tools (pictures and edits, video, voice-overs, Shorts, music, recordings). Settings has the setup line for each.
Bots for creators: new templates for a Shorts channel, a podcast publisher, music for your videos, and a weekly website demo.

0.7.8

Use Claude: Anthropic's models two ways, both on your own account. Your own API key (Cloud models > Anthropic, for example claude-opus-5-5), or your Claude Code: install it, then Settings > Cloud models > Sign in to Claude Code, with your Claude plan or a Console key. Ocadox never sees your sign-in.
Claude Code in agent mode edits files in the chat's folder and asks through Ocadox before commands, the web or anything outside the folder. Its models are marked CLOUD.

0.7.7 October 12, 2026

Music that sounds like music: your chat model turns a short idea into a full description (genre, mood, instruments, texture, tempo), the tempo is set to suit the style, and instrumentals build from an intro to a fade-out.
Best quality: ACE-Step's XL model with its 4B planner, for fuller, more detailed songs. Set it up once in Settings > Image creation > Music: best quality (9.8 GB).
Android: scanning a PC's code works on every phone. When Google's scanner is missing, still downloading or fails, Ocadox reads the code with the camera inside the app instead (Android asks once).

0.7.6 October 12, 2026

Make music (Tools > Make music): songs or background music from a description, instrumental, with lyrics written for you, or your own, 30 seconds to 4 minutes. Made on this PC with ACE-Step 1.5 and saved as an MP3 in the project's Music folder, so there's no copyright strike under your videos. Set it up once in Settings > Image creation > Music.
The agent and bots can make music too (music_create); they ask first.
Your guards cover everything: voice-overs, dubbing and talking characters now wait for the fan guard and run under the power cap, the model's one-off replies (drafts, edit plans, Shorts picks, meeting notes, translations) get quiet pacing, and the agent's media tools wait for the card to cool.

0.7.5 October 12, 2026

Dub a video (Tools > Dub a video, or Dub under an MP4): your model translates what's said and it's spoken in your saved voice where the original was, in 10 languages, with captions and the original sound quietly underneath if you like.
Meeting notes (Tools > Meeting notes): record from the microphone (up to 4 hours) or pick a recording; you get a summary, key points, decisions and action items, saved beside it with the transcript. Action items can go straight onto the board.
Remove the background: under Edit a picture, cut the subject out as a PNG with a see-through background. To erase something, just say what to remove.
Dictate anywhere (Settings > Voice): press Ctrl+Alt+Space in any app, talk, press it again, and what you said is typed where your cursor is.
Android: scanning a PC's code works on phones where Google's scanner failed to open.

0.7.4 October 12, 2026

Long video to Shorts (Tools > Long video to Shorts, or Shorts under an MP4): Whisper writes down what's said, your local model picks the moments that work on their own, and each one is made vertical (9:16) with big word-by-word captions and its title. Tick, rename or nudge them first. Saved beside the video, all on this PC.
The agent and bots can make Shorts too (video_shorts): they read the transcript, pick the moments and save them in the project.

0.7.3 October 12, 2026

Board (Tools > Board): a Trello-style board for each project, kept on this PC. Columns, cards with notes, labels, due dates and checklists, drag and drop. Do it hands a card to the agent, which ticks the checklist and moves it to Done.
Bots can work through a board column one card at a time (Pro): bot routines > When a card is in a board column.
Local servers like Ollama: their replies get the same quiet pacing; LM Studio's models are in the model library (fit labels, Load, Unload, Download) and in the In VRAM panel, and unload when idle.
Model library > Cloud: OpenRouter's models with prices, and your Codex, picked the same way as local ones.

0.7.2 October 12, 2026

Local servers: use models from LM Studio, llama.cpp, vLLM, Jan, KoboldCpp or text-generation-webui, on this PC or your network. Settings > Local servers > Find servers. They're local models: no CLOUD tag, and the fan guard and power cap apply.

0.7.1 October 12, 2026

Edit by transcript (Tools > Edit by transcript, or Transcript under a video): Whisper writes down every word on this PC; click words to cut them, remove filler words and long pauses in one click, add captions that follow the speech, and turn on studio sound.
Overdub: in Fix words, pick what you said and type what you meant; it's re-spoken in your saved voice and put in its place, with the picture held to fit.
Edit video: cut out parts from the middle, and add text that shows between chosen times.

0.7.0 October 12, 2026

Use your Codex: have a ChatGPT Plus or Pro plan? Sign in with ChatGPT in Settings > Cloud models and use OpenAI's Codex as a model, on your plan's limits. Needs Codex installed (npm install -g @openai/codex). In agent mode it works in the chat's folder and asks through Ocadox before anything risky.
Record Ocadox: Tools > Record Ocadox records this window as a video, and Record a demo records while the agent does a task, stopping when it's done. Then Edit it or add a voice-over.

0.6.9 October 12, 2026

Ocadox for Mac (Apple Silicon): the .dmg on the download page. Until it's signed for Mac, open it the first time with right-click > Open.
Git panel (Tools > Git): see your changes and diffs, commit with a message your local model writes, push, pull, switch branches, clone, and open GitHub pull requests.
A calmer top bar: video, picture editing, voice-overs, recording, characters, the Poster and Git are now in one Tools menu.

0.6.8 October 12, 2026

Sign in with OpenRouter: in Settings > Cloud models, press the button and approve in your browser; no key to copy. Optional and off until you choose it.
Pick from OpenRouter's 450+ models, with prices: Claude, GPT, Gemini, Grok, DeepSeek and more. Agent-ready shows the ones that can use the agent's tools.

0.6.7 October 12, 2026

Longer videos: choose 2 seconds to 2 minutes. Long videos are made as 5-second parts, each starting where the last ended, and saved as an MP4. Write one line per scene for a storyboard.
Voice-overs: pick any video, type a script or have your local model draft one, and Ocadox speaks it in your saved voice and adds it to the video, with captions if you like.
Poster: Sign in with Mastodon in your browser, no token to copy, and a button that opens the right page for every other platform.
The browser tool now asks before opening, clicking or typing on a website outside your PC.
Edit pictures by asking ("make it night", "change the sign to SALE") and upscale them 4x: press Edit or Upscale under any picture. Set up once in Settings > Image creation > Picture editing.
Quick video edits: press Edit under a video and describe it ("vertical for TikTok, cut the first 2 seconds, text at the top"). Your local model fills in the trim, shape, music and text.
Record a website as a video: a smooth scroll through any page, desktop or phone sized (command palette > Record a website).
Characters, for AI influencers: a face that stays the same in new photos, a personality, a voice, videos and posts with an AI label. Free has 1 character, Pro 5.
Bots can now make and edit videos, voice-overs, recordings and pictures, and post them. Videos post to Mastodon, Discord and Bluesky.
Images and videos now check how much video memory is free, so they run beside other programs that use the graphics card.
Terms, refunds and open-source licences, linked from Settings > Pro.
Android app: adding a PC no longer closes the app after you allow the camera. Scanning the code now uses Google's scanner, so Ocadox doesn't need camera access at all.

0.6.6 October 11, 2026

Make videos on this PC: /video_create a cat walking through autumn leaves, or Animate under any generated image (Wan 2.2, set up once in Settings > Image creation). Needs a graphics card.
Speech in your voice: add a short sample of your voice in Settings, then /speak and what to say (Qwen3-TTS, on this PC). Only voices that are yours or that you have permission to use.
The Poster: write, schedule and publish posts to Bluesky, Mastodon, Discord and Slack (free: 1 brand, 20 posts a month), plus X and Reddit with Pro (5 brands, 100 posts a month). Bots can draft posts for your OK, and with Pro post on their own where you allow it.

0.6.5 October 11, 2026

Your quiet week: how long Ocadox worked in the last 7 days and how quiet your PC stayed, as a card you can share (in the thermals panel, and offered once a week).
Bots that react: a routine can now run when files change in the bot's folder (or only *.py and the like), or after a new commit. A bot's own edits never set it off.
Share bots as files: Share as a file in the bot editor, From a file when making one. No folders or chats go in the file, and shared routines start switched off.

0.6.4 October 11, 2026

Helpers: the agent can hand a question to a helper with a fresh, empty context, which keeps long tasks on track with local models. One at a time, on the same model, under the same fan guard: the fans stay calm.
Before finishing, the agent checks its own changes with the project's check (cargo check, tsc, go build, Python), or a Check: line in AGENTS.md, and fixes what it finds.
Hooks: an After edit: line in AGENTS.md or OCADOX.md (say, npx prettier --write {file}) runs after every edit.
A browser for the agent: it can open your web app (in Chrome or Edge, hidden), click, type, and read the page and its console errors.
From a terminal: ocadox run "fix the failing tests" --folder path.
Optional cloud models with your own key (OpenRouter, OpenAI, Anthropic and others), off until you add one in Settings, marked CLOUD.

0.6.3 October 11, 2026

Send feedback now includes a problem report (system, graphics, memory, CPU mode, Ollama and the loaded models, the last errors), with no names, folders or chats. You see it before sending.
PCs without a graphics card: the model library says which models fit this PC's memory and which would freeze it, and first-run setup suggests a small one.
Linux: a steadier window on some graphics drivers.
Android app: notifications when a bot finishes or needs your OK even with the app closed (a quiet notification keeps it watching; switch it off on the computers screen), Update installs a new version from inside the app, and images from the full app can be saved to the phone.

0.6.2 October 11, 2026

PCs without a graphics card no longer freeze while a model answers: one core is left for the desktop by default, the reply is redrawn a few times a second instead of on every word, and a model too big for the free memory is refused with a message instead of loading.
The whole desktop app on your phone: on the PC, turn on Full control for your phone (Settings > Other computers) and the Android app opens the full Ocadox, running on the PC. Folders are chosen on the phone from the PC's own.
Android app: a PC that can't be reached says so, with the reason and Try again, instead of an empty list. Lists and chats show Loading, and a message you send appears at once.

0.6.1 October 10, 2026

Image creation, built in: Settings > Image creation downloads an image engine and the Z-Image Turbo model once (6.7 GB, every file checked). Then type /image_create and what you want. Works on NVIDIA, AMD and Intel graphics, or slowly on the processor, with the fan guard and power cap.

0.6.0 October 10, 2026

CPU mode, for PCs without a graphics card that can hold a model (office desktops, built-in Intel or AMD graphics). It turns on by itself when Ollama runs the model on the processor: a smaller context window and no hidden thinking, so replies start sooner.
Quiet on the processor too: a processor limit (all cores, all but one, or half), and the fan guard and the Silent and Balanced profiles watch the processor's temperature.
Processor power while generating: Normal, Reduced or Low, then put back. Windows turns turbo boost off (Low also holds the processor to 70%) with no administrator rights. Linux lowers the processor's own power limit to about 60% or 40%, or exact watts, after a one-time Allow.
Windows: the processor's temperature on more PCs, including without administrator rights.
Linux: installs from the .deb or .rpm now update themselves from inside the app, like Windows and the AppImage.

0.5.5 October 10, 2026

Phone app: much less lag while a bot or chat is replying, especially when the PC is slower. The reply's progress is sent a few times a second instead of on every word, and the phone redraws once per frame.

0.5.4 October 10, 2026

Small screens: clearer Menu and temperature buttons at the top for the sidebar and the thermals, and the window can now be as small as 480 by 400.

0.5.3 October 10, 2026

Smaller screens and windows: the panels make room, then the thermals and the sidebar slide in from buttons at the top. Works with display scaling on Windows and Linux.

0.5.2 October 10, 2026

Try Pro free for 7 days: Settings > Ocadox Pro. No key or card needed.
AMD and Intel graphics: the fan guard and the GPU panel now work with them too (Windows and Linux), not just NVIDIA.
Android: the app tells you when a new version is out.
Ollama cloud models are marked CLOUD; Ocadox explains once what that means before you use one, and Settings > Ollama server can hide them.

0.5.1 October 10, 2026

Phones: the pairing QR code now opens the Ocadox Android app instead of the browser. No app? A button shows the browser code.
Phones: a Chats tab next to Bots. Read your PC's chats, or start a quick chat with its model (no tools); it's saved in the PC's chat list.
Phones: turn a bot's routines on or off, or run one now.

0.5.0 October 10, 2026

Bot templates: start a new bot as a test watcher, code reviewer, dependency checker, weekly summary or folder tidier.
Pro: your bots on your phone. Get the Ocadox Android app from the download page, then Settings > Other computers > Pair a computer or phone, and scan the QR code. Add all your Windows and Linux PCs; chat, stop and approve from the phone.
When a bot finishes, the notification shows the start of its answer. Routines can also save each result as a file.
Ocadox starts Ollama by itself when it isn't running, so routines work right after a restart.

0.4.1 October 9, 2026

Models: drag the model cards up and down to put them in the order you like. The order is remembered.

0.4.0 October 9, 2026

Voice: click the microphone by the message box and talk. Your words are written in for you, on this computer, with Whisper.
Settings > Voice: choose the speech model (Small by default) and language.
Pro: the agent can transcribe audio and video files in your project, with timestamps.

0.3.1 October 9, 2026

Bots stay free. Routines, Always on and Other computers are now part of Ocadox Pro.

0.3.0 October 9, 2026

Bots: agents you come back to, each with its own avatar, chat, folder, model and instructions. Their avatars show what they're doing.
Bots keep working, and keep their results, with the window closed.
Routines: schedule a bot (daily, weekdays, weekly or hourly). It can even wake the PC from sleep.
Settings > Always on: start Ocadox with Windows, quietly in the tray.
Settings > Other computers: use an always-on PC's bots from your laptop, securely paired.

0.2.1 October 8, 2026

Fixed: after the rename, the agent could not find its project folder. Your Hush Workspace is now Ocadox Workspace, files included.
A failed step opens by itself, so you can see why without clicking.

0.2.0 October 8, 2026

Hush is now Ocadox. Your chats, settings and Pro license carry over.
Project instructions now live in OCADOX.md (/init writes one for you).

0.1.1 October 8, 2026

Ocadox Pro is on sale: buy a lifetime license in Settings > Ocadox Pro and activate it with the key from your receipt.

0.1.0 October 8, 2026

A guided first run: the right model for your GPU, your project folder, and a first task.
The agent notices when it is stuck in a loop and changes course or stops.
Share how quiet a run was as an image.
Ctrl+K command palette, Ctrl+N new chat, Esc to stop, and a light theme.
/image_create makes images on this PC: built in (set up once in Settings), or with your own Forge server.
Background commands with a live terminal panel, and approving each edit with a diff.
Plan mode, OCADOX.md project instructions, images for vision models, PDFs and Word files.