What's in the free download.

The short answer first. Then the whole picture.

A big question gets a straight answer in a few seconds: two to four sentences with only what's certain. Then an outline appears, and each part is researched, written and checked by its own AI, all at the same time. A part shows up once it passes its check, and nothing you've already read is changed.

• One AI for a quick question, up to eight for a report
• The whole answer takes about as long as its slowest part
• Each part keeps its own searches and sources, open them to see how it was made
• Off, Auto or Always in Settings, or for one message from the send arrow

Two sliders. Mosaik tunes the rest.

Cost sets the most an answer may cost, from Free to No limit. The second slider sits between faster and smarter. For each message, Mosaik then picks the model, how long it thinks, whether it searches, how many AIs work on it and how the answer is laid out.

• Free models are the default
• A model that's known to do a job badly never gets that job
• Each answer's details say what Smart picked, and why
• Your daily and monthly limits still come first
• Manual mode puts every setting in your hands, for one message, one chat or all of them

Fast, and honest about what it's doing.

Answers start streaming straight away. Above each one, a single line says what the model is doing (connecting, thinking, searching, reading a page), then sums up the turn when it's done: "Searched the web and thought for 14s". Open that line to see every step in order.

• Each chat can have its own model, instructions, temperature and length limit
• Keep typing while it answers: your next message waits its turn
• Edit any question or regenerate any answer, or branch into a new chat from any point
• Scroll up while an answer is being written and the page stays where you put it
• Half-typed messages wait in their own chat, and in New Chat, until you come back

Three models, one conversation.

Put two or three models over the message box and talk to all of them at once. Their answers sit side by side in the chat, each with how long it took and what it cost. Keep the one you like and the chat carries on with that model. The others stay as versions.

• From any answer's menu, the send button or the model browser
• Nothing is asked until you send
• Each model keeps its own side of the conversation
• Scroll a column while it's still being written

Ask your spreadsheet a question.

Attach a spreadsheet or a CSV and ask what's in it. The model runs Python over the file itself, so totals, averages and trends are worked out, not guessed. Ask for a chart or a cleaned copy and it comes back as a file in the answer.

• CSV, Excel and OpenDocument, every sheet
• Bar, line, scatter, pie and histogram charts
• Your files are only read, in a locked-down sandbox

Every chat, one keystroke away.

Ctrl K searches every conversation by title and by what was said, and shows the passage that matched. Ctrl P runs any command or switches model, with the shortcut beside each one, and short commands work in the message box itself: /new, /search off, /think high.

• Chats name themselves after the first answer, and never rename one you named
• Find in the open chat with Ctrl Shift F
• Pin chats to the top and file them in folders
• Ctrl-click to pin, file or delete several at once
• Export to Markdown or JSON, reasoning included

It remembers what you tell it, and shows you.

Mention your job, your tools or how you like answers, and every chat knows it from then on. When a reply saves something, a line under it says what, with Undo. Settings lists everything Mosaik remembers, to change or delete. It all stays on your computer.

• It can look back at your past chats when you mention one
• Temporary chats aren't kept, listed or remembered, unless you choose to keep one

Answers read aloud.

Any answer can be read aloud from its ⋮ menu, with your computer's voice for free or a speech model for a more natural one. It reads the words and skips code and files.

• Your computer's voice stays on your computer
• A bar under the answer has Stop

Share a chat as one file.

Save a conversation as a web page or a PDF that looks the way it does in Mosaik, charts and all. It's one file on your computer. Nothing is uploaded, so you decide who sees it.

• Choose whether reasoning, pictures, files and sources go in
• The page follows the reader's light or dark setting

Ask the same thing faster.

Save what you ask often in the prompt library, with blanks like {topic} for the part that changes. Type / in the message box and your prompts come first. Pick one, fill in the blanks, and read it over before it goes.

• A blank can start filled in: {tone:friendly}
• Keep chat instructions in the same library

Give it a job, not just a question.

Ask for a book outline with a file per chapter, a script that has to run, or notes to sort, and it works through it step by step in a folder of the chat's own, saying what it's doing as it goes. A quick question stays quick: it reads the instructions for its tools only when a task needs them.

• Up to 100 steps a message, then Keep going
• Open another chat and this one keeps working
• Reminders for you, or for itself to follow up later

Laid out like something you'd keep.

A reply can highlight the one line that matters, set dates on a timeline, put key numbers in large type, or compare options side by side with the best one marked. It streams as fast as plain text, in the conversation's own fonts and colours. Chit-chat stays chit-chat.

• Off, Light, Auto or Rich, in Settings › Answers › Styled answers
• Text size, spacing and reading width in the same place

Designed beside the chat.

Ask for a website, an app screen, a poster or a small game and it opens in its own window next to the conversation once it's done. Keep chatting to change it. Every change is a new version, a click apart.

• Switch between the preview and its HTML
• Copy it, or save it as a web page

Smart picks, or pick your own.

Smart ranks models by their published scores and prices, so a better or cheaper model is used as soon as it appears. Or choose one model for a chat, keep saved setups in the model menu, and browse every model by price, context size and what a model can do.

• Change the model for one chat from the button in the message box
• Reasoning effort per model: Auto, Off, Low, Medium or High
• Ask OpenRouter for the fastest or the cheapest provider
• Test checks a model can use Mosaik's tools before you rely on it

Say what you want changed.

"Make it cheaper." "Use Claude for everything." "Stop searching the web." Ask in any chat, and the assistant reads your settings and proposes the changes on a card under its answer. Untick any you don't want, then Apply, and Undo puts them back. It never sees your key.

• A setup guide on first launch: key, models, limit, done
• Tips that appear once, next to the feature they're about

Further off: everything else you do with AI.

The Mosaics tab is already in the app, showing what's coming: tools for code, writing, art and research that you can use on their own, and mosaics, projects that keep all of it together with one AI in charge. It will be an optional paid add-on. Chat stays free.

The roadmap · What Mosaics will do