Where your work lives, and who can see it.

On your own computer. The desktop app keeps project files, conversations and logs locally, and you can use existing folders on your computer as projects. None of this is stored on our servers.

Importing a file doesn’t upload it anywhere. Content is only sent to a model when you ask the AI to work on something, and only what that task needs. Automatic AI tagging on import is off by default.

On iPhone, the originals stay inside the app and are left out of iCloud and other system backups. When you send them to your computer through the cloud, the relay server only holds them briefly: as soon as your computer confirms receipt, the cloud copy is deleted. All that remains is a delivery record with no file content.

If a file is never collected, or a deletion hits a snag, a daily cleanup removes anything that has been there for 30 days. The original on your phone and the copy on your computer are yours to manage; deleting one doesn’t affect the other.

Same idea, different time limits: files sent from your computer to your phone wait in the cloud for 7 days, and files sent from your phone to your computer for 30 days, then they are cleaned up automatically. The relay is for delivery, not backup, so save anything important on the receiving device.

It depends on what they connect to. When the Office add-in connects to the desktop app on your computer, data stays local; when it connects to a cloud service, conversations and sync data are stored on that server. Conversations with the Skill drafting assistant on the website are stored on our server too.

Team and firm memory is shared between members, so it lives on our server. If you set up remote sync such as Git, memory is also synced to the repository you choose. Account, order and usage records are kept on our server for billing.

Because your work lives on your computer, you’re in control of the backup. Before moving or uninstalling, copy your project files, conversation data and personal and project memory. Signing in on a new computer won’t bring back what was on the old one.

Deleting local files, shared cloud content, your website account and external sync copies are separate steps. The phone relay isn’t a backup. Please check diagnostic logs for case content before sending them to us.

No. AI WorkDeck uses existing AI models, and we never train models on your material. “Memory” is simply a readable, editable note of your preferences, case facts and working experience that is handed to the model as reference next time. The model itself doesn’t change.

Model requests go through OpenRouter to each model provider, which handles them under its own privacy policy; see the link below. For highly sensitive material, you can switch to a local model or consider self-hosting.

Personal and project memory is kept in a local database on your computer, where you can view, edit or delete it at any time. If you turn on Git sync, it is also synced to the repository you set up.

Team and firm memory is shared, so it lives on our server and only members of that organization can read it. Team admins manage team memory, and only managers in the head-office team can write to firm memory. Before adding something to shared memory, make sure it’s right for everyone who can see it.

Find that memory and edit or delete it. Keep anything specific to one case in project memory rather than team or firm memory. Memory is only a reference for the model and it can still misread it, so double-check key facts and rules before relying on them.

Yes. Community Edition can be deployed on your firm’s own server with a local model, so storage and processing happen in-house.

To keep everything truly in-house, switching to a local model isn’t enough on its own: other online features also need to be turned off or replaced with in-house services, such as cloud OCR, transcription, web search, the phone cloud relay, shared memory on our website and remote sync. Local models also vary in tool use, image understanding and other capabilities, so each needs testing and may not match the default cloud setup exactly.

If this is what you need, get in touch with your team size, the model you plan to use, the features you want to keep and your network requirements, and we’ll work out the setup together.

Only what the task needs: your question, the relevant conversation, file content you provide or the AI reads for that task, and any memory it uses. Your whole project isn’t sent by default.

When you chat with AI in the desktop app, the request goes straight from your computer to OpenRouter, which passes it to the model provider. It doesn’t pass through our servers, and we can’t see your conversations; our website only handles your account and usage billing. Other add-ins and online tools work according to their own settings.

Desktop conversations and logs are saved only on your own computer, not on our servers. Content is stored on our servers only for the cloud features you choose to use: the cloud Office add-in, the Skill drafting assistant, shared team memory and the phone relay, each described above.

Whether a model provider keeps requests depends on its own policy; see OpenRouter’s explanation below.

Keep two things apart: organizing files locally doesn’t send their content anywhere, while asking the AI or an online tool to work on something sends the relevant content to that service. Before you start, be clear about what will be sent and where.

A common approach is to share only the excerpts the task needs and remove names, ID numbers and other details you don’t need. If a client or your firm requires that material never leaves, don’t give it to a cloud model; use a local model or a self-hosted setup instead.

No. Joining a team doesn’t share the projects or files on your computer. Colleagues can only see what you add to shared team or firm memory, plus the usage information the team dashboard shows based on roles and settings. Whether project names are shown is a separate setting.

Invite members and grant admin rights only as needed, and keep one client’s confidential details out of memory the whole team can read.

Not without checking. Go back to the original source and confirm the jurisdiction, publication and effective dates, the version, case number, court, and whether the quotation really supports the conclusion. Models can invent sources or misapply real ones.

Treat AI output as research material and first drafts. Findings of fact, legal analysis and the final work product still need the lawyer handling the matter to sign off.

Check names, amounts, dates, negations, clause numbers, attachments and tables that span pages. Recognition and summaries can miss details, and the model may not have seen the whole document.

Keep the original and compare versions. Review any change that affects rights or obligations, along with quotations and conclusions, by hand; a summary is no substitute for the underlying evidence.

Find out which files it reads, which services it contacts and what it can do. Third-party tools handle data under their own rules, and documents or web pages can contain hidden instructions that mislead a model.

Give trusted tools only the access they need, and look over important writes, deletions or outside submissions before and after they run. Never post client material, keys or passwords in the public community when asking for help.

The mainland site, aiworkdeck.com, runs separately from the overseas sites; accounts and online data aren’t shared. workdeck.ai and aiworkdeck.us are the same overseas system and share user data.

Connect the desktop app to the site where you registered and topped up. Switching the website language doesn’t move your account, balance or local projects.