Start here to install Cuttlefish, connect the models you want, set things up for you and your team, and keep control of your data. Every guide below is a real path, not a list of links — pick the one that fits what you're trying to do.
Cuttlefish is a desktop app you install once and open like any other program. Download it, open it, sign in, and connect a model — then you're chatting, building, and browsing. New here? Walk it end to end and try the things worth doing first.
Get the version that matches your computer — Windows, macOS, or Linux — and install it the usual way. The website highlights the right download for the machine you're on, so the obvious button is almost always the one you want. Updates arrive on their own afterward; you check for them from inside Settings whenever you like.
After you sign in through your organization’s approved access path, Home helps you start and supervise work. Home, Workspace, Browser, Apps, and Settings are the main navigation. Independent Companions, Living Surfaces, Desktop Presence, Working Groups, and Stage keep the relevant work beside you. Switching views does not end an assistant’s authorized task.
Sign in with your work account and follow your organization’s access instructions. To connect your phone, open Settings → Mobile on the desktop and follow its pairing instructions. Scan the displayed QR code or enter the pairing details in the installed Cuttlefish phone app. Available remote work and approvals follow your account permissions and the paired desktop’s availability.
Use a model connection approved by your organization, or add an allowed provider in Settings → Connections. For a cloud provider, use its API credentials; for a model on your computer, choose its compatible local endpoint. Test the connection before using it, then select an available model where you work.
Beyond the basics, a few minutes of setup makes the day-to-day smoother: build and keep your own apps, connect the tools you already use, and find your organization's plan and usage in one place.
Describe what you want in plain words and Cuttlefish builds it on a canvas — a document, a page, a dashboard, a small tool — previewed live in a walled-off space you can experiment in freely. When something you built is worth keeping, turn it into a real app that lives in your Apps library and opens like any other tool, with the ability to roll back to an earlier version.
Link the services you already rely on — your code, your cloud accounts, and more — with guided setup from the Apps tab. Pair local tools to a private on-device connection, or connect remote tools through a governed one. Connecting a tool doesn't hand anything over; Cuttlefish only does what you've allowed, and you can change or remove a connection any time.
Your organization’s plan and usage are in Settings → Billing. Signing in gives you access to account history and organization resources allowed for that device. Files, browser profiles, device credentials, and models on another computer stay on that computer; phone pairing gives access to the desktop features your organization permits.
Cuttlefish for teams is the same product with the controls a workplace needs layered on top — not a separate tool. Admins claim a domain, set up single sign-on, and invite people, all with clear visibility and proof of every action. Start at the enterprise setup guide.
Connect your organization's domain so the right people land in the right place when they sign in, and so admins can govern who gets access. This is the first step in standing up a team workspace, and it sets the boundary everything else sits inside. Walk it through on the enterprise setup page.
Wire Cuttlefish to your existing identity provider so people sign in with the credentials they already use, and access follows your directory. It's the foundation for clean onboarding and offboarding across a team. See how it fits the broader admin picture under Enterprise.
Send invitations, set who can do what, and keep an audit trail of actions across the workspace. People share work, models, browsers, files, and internal systems in one place, while admins keep visibility and proof. A team can start small and grow on the exact same tool.
Most AI apps give you one company's model in their box. Cuttlefish works the other way: bring the models you like, pick the one that fits each task, and change your mind whenever you want.
Connect supported provider APIs, including OpenAI, Anthropic, Azure OpenAI, xAI, DeepSeek, Mistral, and Groq, through Settings → Connections when your organization permits them. A ChatGPT or Claude consumer subscription does not supply API credentials or API usage. Available models depend on your provider account, endpoint, and organization policy.
Use Ollama, LM Studio, or a compatible model endpoint on your computer for on-device inference. Prompts and responses are processed there rather than by an outside AI provider. Conversation history and memories are still saved to your Cuttlefish account as described in the Privacy Policy. A model on your network or in a private cloud processes requests at that endpoint.
The model picker sits right where you work. Start a chat with one model, then switch to another for the next reply without losing your place — the thread keeps going, only the mind behind it changes. Use a cloud model for one task and a local one for the next, on the same day, in the same workspace.
Choose approved models and systems, keep work within your organization’s permissions, and review actions and evidence. Files, hosted account history, model requests, and saved action evidence have different data paths; the Privacy Policy explains them.
Nothing goes to an AI provider's servers unless you choose a model that runs there. Keys for models on your own machine are kept in your computer's secure store, cloud provider keys are stored encrypted in our credential service, and your conversation history is saved to your account. Read the full picture on the privacy page.
Reading and preparation run within the permissions your organization has allowed. When an action needs additional approval, Cuttlefish shows its target and proposed change for an authorized person to review. Existing grants and organization rules determine which actions can proceed; blocked actions remain unavailable. See the trust center.
Cuttlefish keeps a plain, tamper-evident record of what it used, what it changed, who approved it, and where the result went — readable in clear terms whenever you want to check. For the security posture behind it, see the security review.
Most issues are quick to sort out. Report a problem with redacted details straight from the app, check whether a hiccup is on our side, or reach a person for sign-in, billing, or setup help.
Open Settings → Help & Support to prepare a diagnostic report. Review the report before sharing it and remove any account, customer, file, or task details you do not want to send. Send it through the support page with a description of the problem.
If something feels slow or unavailable, check whether it's a known issue on our side before digging in. The status page lists the parts of the service and where to follow updates, so you can tell a local snag from a wider one.
For failed sign-in, expired invites, billing questions, or enterprise setup problems, the support page routes you to the right mailbox. Evaluating Cuttlefish for a team? Start a conversation through contact sales.
Quick answers for anyone arriving from ChatGPT, Claude, or Gemini. These cover the things people wonder about before they download.
Those apps give you their model, in their window, on their servers. Cuttlefish gives you your workspace: you pick any model and switch it mid-conversation, you can run one entirely on your own machine, and it works with a real built-in browser and your actual files and apps — not a sealed chat box. It can also keep working on lasting tasks and report back, instead of waiting for your next message.
No. You talk to Cuttlefish in plain language, the same way you would with any AI assistant. When it builds something for you — a document, a page, a small tool — you describe what you want and watch it appear. You never have to read or write code.
Yes, for AI inference. A model running on your computer processes prompts and responses there without sending them to an outside AI provider. Cuttlefish still saves conversation history and memories to your account. Local files stay on your device, while excerpts or images used in a task can become part of its conversation or action evidence. See the Privacy Policy for storage and erasure details.
Install Cuttlefish, connect a model, and start your first conversation — or talk to us about bringing it to a team.
