Security, privacy, and trust
How Sai handles your data, isolates workspaces, protects credentials, uses AI models, controls access, and deletes what you remove.
What data does Sai store?
Sai collects and stores only the data needed to run autonomous workflows, keep session history, and preserve what you create.
| Data | What is stored | Where it lives | Retention |
|---|---|---|---|
| Account and profile | Email address, authentication tokens, profile preferences | Encrypted cloud database | For the life of the account |
| Task instructions and run logs | Prompts, tool calls, console logs, intermediate reasoning checkpoints | Cloud task logs | Until you delete them or remove the account |
| Workspace files and deliverables | Documents, datasets and exports written to artifacts/ | Your cloud file library | Controlled by you; kept until deleted |
| Website and app credentials | Usernames, passwords, API keys, TOTP seeds | Sai Vault, with hardware-backed encryption | Managed by you; deleted immediately on removal |
| Screen data during a run | Screenshots, DOM trees, accessibility trees | Temporary memory during the active run | Flushed between steps; never stored permanently |
Who can see your tasks and data?
Access to workspaces, files and runs is separated between automated agents, Simular staff, and members of your team.
Isolated workspaces
Every user works in a logically isolated workspace. Agents cannot inspect, query or run commands against another account's environment or files.
Personal and team workspaces
- Personal tasks are visible only to the account owner.
- Shared team workspaces show designated routines, canvases and deliverables to invited members according to their role: Admin, Member or Viewer.
- Deliverables stay private unless you share them with a workspace link or public URL.
Simular staff
Simular staff cannot view active desktops, read stored passwords or open private files unless you request and authorize a support intervention. Every operator intervention is audited and logged.
Does Sai train AI models on your data?
No. Sai connects to frontier model providers, including Anthropic and OpenAI, through enterprise API agreements under which your prompts, screen observations, documents and code are never used to train, retrain or improve foundation models.
Passwords stay out of model context
Website passwords and TOTP codes are handled through Sai Vault or isolated input bridges. Opaque secret handles put credentials straight into sign-in fields, so the raw values never enter a model's context window or your conversation history.
Bring your own key or model
Enterprise and advanced plans can route reasoning through your own API keys or self-hosted models, so prompts never leave your network perimeter.
How Sai prevents unwanted actions
Sai combines infrastructure security with human-in-the-loop approval for anything with consequences.
Approval tiers
- Safe actions run automatically: read-only checks, DOM inspection, visiting public URLs, local calculations.
- Guarded actions need your confirmation: sending email, publishing posts, completing payments, changing CRM records, deleting files.
- Irreversible actions wait in the Needs You queue until an authorized person approves or denies them.
Encryption
- In transit: network traffic, remote desktop streaming and API calls use TLS 1.3.
- At rest: file storage, databases and backups use AES-256.
- Vault: integration tokens and passwords use envelope encryption with hardware security modules (HSM).
How to delete your data
You own your data and can remove it at any time.
Delete individual items
- Files: remove any document, image or dataset in artifacts/ from the file manager or through the API. Deletion applies to active storage and backups.
- Credentials: removing a login or key revokes it and purges it from the vault immediately.
- History: clear conversations, reset sessions or prune run history whenever you want.
Delete your account
Deleting your account permanently removes your workspaces, file library, connected tokens, routines and logs. The cloud computer disks assigned to you are securely wiped and destroyed.
Frequently asked questions
Does Sai train AI models on my data?
No. Sai uses model providers under enterprise agreements that exclude your prompts, screen observations, documents and code from model training.
Can Simular staff see my screen or passwords?
No, unless you request and authorize a support intervention. Interventions are audited and logged.
Which actions need my approval?
Anything with consequences: sending email, publishing posts, completing payments, changing CRM records and deleting files. These wait in the Needs You queue until you approve or deny them.
How are my passwords protected?
Credentials are stored in Sai Vault with hardware-backed envelope encryption and are entered into sign-in fields through opaque handles, so they never appear in a model prompt.
Can I use my own model or API key?
Yes, on Enterprise and advanced plans. Reasoning can run through your own API keys or self-hosted models.
How do I delete my data?
Delete files, credentials or history individually at any time, or delete your account to permanently remove everything, including the disks of your cloud computers.