No sandbox
Helpers run like ordinary programs under your Windows account. The project folder is where they start, not a fence.
What Breezable can do
Breezable keeps your chats, your AI workers, the work they make and the things worth remembering in one Windows app. Here's how the pieces fit together. Screenshots show the real interface with sample data.
Otto runs the team
Otto is your coordinator. Chat with Otto like you would with any AI, or hand over a bigger goal and Otto can start a team of Claude and Codex workers running at the same time.
- See the whole teamSee every worker, what it's doing, and which AI model and effort it uses.
- Read any worker's conversationOpen a worker for its conversation, steps, live output and files. What a worker reports is its own account, and Breezable labels it that way.
- Questions in one placeWhen an AI needs you, the question shows up in one list. Pick a suggested answer or write your own.
Your choice of AI, every time
A quick question or a big plan, you decide which AI model does the thinking and how hard it works at it.
- Pick the model for each chatChoose which AI answers and how much effort it puts in. Each chat remembers its own choice. Available models depend on your account.
- Drafts stay where you left themSwitch between chats without mixing up half-written messages or attachments.
- See how much room is leftA context meter shows how much of the conversation the AI can still hold before older parts get compacted.
Results open next to the chat
Read the work in the same place you talk about it. Files, the browser, the terminal and the Brain open as tabs next to the conversation.
- Page view or code viewSwitch an HTML file between the rendered page and its source. Markdown renders too.
- Room to focusKeep the chat and the tab side by side, or expand the tab to the whole window.
- Search inside a filePress Ctrl+F to find text in the file you're reading.
Otto Main Brain remembers what matters
A searchable memory on your computer, so later work can find what earlier work learned. It's organised as a garden of topics, and it's new as of October 2026.
- Browse or search itTopics hold lessons, facts, notes, web captures, skills and more. Search everything, or open the map view.
- Tend itPin, move, mark as outdated or forget items. A journal, pruning and restore keep it manageable.
- You choose the automatic partsSome things are captured automatically, and a switch controls saving lessons after work. Unreviewed items stay out of research unless you opt in.
Is everything in the Brain correct?
Not necessarily. It holds what was saved, and that can be wrong or out of date. Check anything important against its source.
Give a task only what it needs
Search your sources, pick the passages that matter, and build a focused package instead of sending everything at once.
- Find the right passagesSearch your sources, then filter them by kind, area, tag or review status.
- Pick up to eightAdd the best passages to a tray. Each one keeps its source attached.
- Set a size limitChoose an approximate token budget, and see what got left out and why.
Why does sending less help?
AI services count how much text they read, and limits or costs follow from that. Relevant information also tends to give better answers. We don't publish a savings figure, because it depends on the job.
The right model for the work
Different work needs different amounts of thinking. Breezable lets you decide what runs where, or leave it on Auto.
- By work type and difficultyA table sets which model and effort handle each kind of work at each difficulty.
- Cost-Effective, Performance or Highest QualityChoose how Otto balances speed, quality and your usage limits. These are relative choices, not measured savings.
- Your own accountsWorks through the official Claude Code and Codex command-line tools. Which models you can use depends on your account.
What helpers can do on your computer
Breezable's helpers do real work on your machine, and you should know how far that goes before you start.
They don't ask every time
Helpers start with their normal permission questions turned off. Expect them to change files and run commands without asking first.
Questions are for decisions
Workers can ask you things when a decision is needed. Answering gives them information. It isn't a safety check on what they do.
Lower what's at stake
Use a separate Windows account, back up important files, and only open projects you trust. This reduces the risk. It doesn't remove it.
Make it work your way
The details are there when you want them, and out of the way when you don't.
Settings the AI can change
Tell the AI what you want changed. You see each change first, can apply it, and can undo it. Secret values have no undo.
Voice dictation
Dictation runs on your computer by default, using a speech model downloaded on first use. Cloud dictation through OpenAI is only used if you choose it.
See your usage
A usage view shows tokens by time and tool, and an estimated API value. That figure is an estimate, not a bill.
Comfortable to use
Change the theme, accent color, text size, chat width and motion. Rebind the keyboard shortcuts.
Say what worked
Mark finished work as worked or not, or skip it. Your review can feed what Breezable learns for later jobs.
Debate rooms
Let Claude and Codex argue a question for the rounds you choose, then write a verdict.
Logins with your say
When the AI needs a saved login, it asks: allow once, always on this site, or deny.
Test VMs, for technical users
AIs can run work inside disposable VirtualBox virtual machines. This needs VirtualBox and test images, and it's meant for people comfortable with that.
See it on your own work
The public installer is coming soon. Until then, see what you'll need.