Trust
How we protect your data.
Your Talent works inside your tools, with your permission, under your approvals. Your private conversations stay private. Here is exactly how, in plain English: what we collect, who can see it, where it lives, and how to get it back.
The short version
Private by default
Your chats, memory, runs, connections, and approvals are readable by you alone. No admin setting opens them. This is enforced in code, not in a policy paragraph.
Admins see the bill, not the work
Workspace admins can see what a job cost and that a teammate has a Talent. They never see what was said or produced.
Nothing acts without permission
Sending mail, changing a calendar, or publishing something can require your approval first. Money movement and irreversible actions always ask.
Shared knowledge needs a human yes
What enters your company's shared knowledge is proposed by the teammate and approved by a person before it spreads. Private memory never joins it automatically.
Credentials stay unreadable
Connected tools are reached with per-person credentials that are encrypted, scoped to one person and one Talent, and never displayed back after you connect.
You can leave with your data
Download your documents and artifacts at any time. Deletion and retention terms are spelled out below.
We do not train AI models on your work, and we do not let our model providers train on it. Your work is not a training set.
Who can see what
Roles exist to run the company, not to read people's work.
| What | Who can see it |
|---|---|
| Your private conversations with your Talent | Only you |
| Your personal memory, the context your Talent keeps about your work | Only you |
| Your private runs, approvals, and suggestions | Only you |
| What a job cost, when it ran, and which model it used | Workspace admins (billing only) |
| That a teammate has a Talent, its name and status | Workspace admins (identity only) |
| Company knowledge and shared workflows | The team, after a human approves the promotion |
There is no admin switch that opens a member's private chats. The queries that read private data are owner-scoped in the database itself, so a misconfigured role cannot widen them. Personal data and shared company knowledge are separate systems with a human approval in between. The full legal version lives in our Privacy Policy.
Controls, approvals, and limits
Your Talent can read, draft, and prepare work freely; those actions are reversible. Actions that leave your workspace are treated differently, and the system applies controls before they happen.
Approval gates
Sending an email, posting as you, writing to your calendar, or publishing a document can park and wait for your yes, your no, or a yes for this conversation only.
The tier that always asks
Money movement, irreversible actions, and calendar writes always ask. No setting turns those off.
Allowlists
Not every tool is reachable. Each connector exposes a deliberately small set of actions, and the platform enforces it.
Budgets and brakes
Credits and turn budgets cap what one task can consume, and a loop guard stops a stuck job instead of letting it run all night.
Receipts
Every action leaves a record: what ran, what it touched, and what it cost. You can always see why something happened.
Grounded, not guessing
If a connection is missing or broken, your Talent says so instead of improvising. It reaches only the tools you connected and authorized.
Data privacy and sovereignty
Where your data lives, who legally controls it, and the commitments that apply on every plan.
Your organization is in control
Your organization is the data controller for workspace data. TalentOS acts as a processor and handles data only on your instructions.
Stored and processed in the United States
Workspace data is stored and processed in the United States on the providers named below. For customers in the EEA or UK, transfers are covered by Standard Contractual Clauses.
Encrypted in transit and at rest
Data is encrypted in transit, and stored credentials are encrypted at rest.
Limited internal access
Access to customer data by our team is limited to what you ask us to do, and to what security, abuse prevention, or the law requires.
No sale, no advertising, no training
We do not sell customer data or use it for advertising, and your content is not used to train models, ours or our providers'.
Google data commitments
For data from connected Google services, we comply with the Google API Services User Data Policy, including its Limited Use requirements.
Where your data lives
We run on established infrastructure, and we name every company that touches your data. This list is kept current, and Enterprise customers hear from us before a material change.
| Company | What it does for us | What it can touch |
|---|---|---|
| Clerk | Sign-in, sessions, and workspace membership | Account identity |
| Convex | Application database and backend | Your workspace data at rest |
| Railway | Hosting and delivery of the web app | Serves the application; no separate data store |
| Stripe | Payments and billing for paid plans | Billing details; never your conversations or documents |
| Anthropic, OpenAI, and Google | AI models for drafting, research, analysis, and summaries (accessed through OpenRouter) | The content needed for the task you asked for |
| Groq | Voice transcription for voice notes | The audio you record, at the moment you record it |
| Firecrawl | Fetching public web pages for research | Public URLs only; never password-protected pages |
| E2B | The isolated sandbox that runs your Talent's work | Files for the task at hand, in a per-session sandbox |
| Fly.io | Runs our iMessage bridge | iMessage messages you send to your Talent |
| Sentry | Error monitoring so we can fix breakage fast | Diagnostic and error data |
| PostHog | Product analytics | Usage events |
| Google Fonts | Typography delivery | Nothing about you |
Customer conversations and documents are processed to do the work you asked for. They are never sold, and they are not used to train models. Slack, Google (Gmail, Calendar, Docs, Sheets), Microsoft, GitHub, Notion, and the other tools you connect are your accounts, not ours. You connect them, you choose what each can reach, and you can disconnect them at any time.
Retention, export, and deletion
Retention
We keep workspace data for as long as the workspace is active. When a member leaves, their private conversations and personal memory are removed with the account. Workflows and knowledge they published to the workspace stay, because those belong to the organization. When a workspace closes, we delete or anonymize its data within 90 days, except where the law requires us to keep records.
Export
Your documents and artifacts can be downloaded from your Library at any time. For a full export of your workspace data, email support@talentos.so and we will prepare it.
Deletion
Ask us to delete your account or workspace at any time. We delete what we are not legally required to keep and confirm in writing when it is done.
Compliance, stated precisely
We would rather be precise than impressive.
SOC 2
Not done yet. It is on our roadmap, and we will publish it when it is real. We will never show a badge we have not earned.
DPA
Available today for Enterprise agreements. Ask us and we will send ours.
Security reviews
Enterprise includes a security review with our team.
Subprocessors
Every company that processes data for you is named on this page, and Enterprise customers are notified before a material change.
Report a vulnerability
If you find a security issue, tell us at support@talentos.so. We will confirm receipt within one business day, keep you posted, and credit you if you would like. Please do not test against other customers' workspaces or run automated scans that degrade the service.
Questions this page did not answer? Ask us directly.
A founder reads every message. Bring your compliance reviewer.
Last updated: September 17, 2026 · Privacy Policy · Terms