Separate for every customer
Each organization’s content is held in its own database area, and workspaces are kept apart inside it. The separation is enforced by the database itself, not only by the app.
STAYCHARTED AMT / PRODUCT
Your examples train only your own models. They’re kept in a database area that belongs to your organization, encrypted, hosted on AWS in the United States, and deleted on a schedule you can see below. StayCharted staff cannot sign in as you.
YOUR DATA, THROUGH ITS LIFECYCLE
Know who can use a model, what masking changes, and what remains after an upload is deleted.
See retention periods ↓Each organization’s content is held in its own database area, and workspaces are kept apart inside it. The separation is enforced by the database itself, not only by the app.
Public connections use TLS 1.2 or 1.3, with HSTS. The database, its backups and stored files are encrypted at rest, and public access to content storage is blocked.
Uploaded picture files are kept as sent for the applicable retention window. AMT’s model and Review Queue copies are generated from pixels only, without location, camera or other hidden metadata. Originals in your storage are never changed.
Your examples train models in your workspace only. The shared pretrained AI does not learn from your uploads.
Uploaded files are scanned for malware and verified against their real contents. Data files with macros are refused. Spreadsheet formulas are neutralized when you download results.
Our own models train and answer inside our AWS account. An outside AI service receives your data only if you connect an assistant or AI platform and use it.
The database runs in two AWS Availability Zones, with encrypted backups and point-in-time restore. Restores are tested.
Assign Owner, Builder, Reviewer and Viewer roles. Keep access appropriate to the work each person does.
Give integrations their own API keys and revoke keys you no longer need. Protect them in your own systems.
The Owner allows assistants, each model has an access switch, and the connection acts with the connecting person’s role.
Assistant providers receive the information read through a connection under your own agreement with them. Activity records changes made through the assistant.
Model downloads: Builders can download a trained model from Manage; what’s in it is set out in the Terms.
PREPARE BEFORE TRAINING
AMT checks supported patterns such as email addresses, phone numbers, payment card numbers, and some secrets. Mask detected values, exclude rows, or keep them. Models trained with masking apply it to new text.
The checks do not detect every name, address, or sensitive free-text detail. Deleting an original upload also does not remove what an existing model learned; correct the dataset and retrain, or delete the model.
How personal-information checks work →
RETENTION AND DELETION
| Data | Live Service retention |
|---|---|
| Original text uploads and picture ZIPs | 90 days after upload, unless deleted sooner. Derived data and trained models remain. |
| Filled files | 24 hours after first download, or 7 days if never downloaded. |
| Files uploaded to fill but never run | 7 days. |
| Rows read from Snowflake or Databricks | As an uploaded file: training rows as an original upload, rows to fill as a file to fill. The results table in your warehouse is yours to keep or drop. |
| Try-box questions | 30 days, unless feedback was given. |
| Example rows in an older validation report | 14 days after a newer version replaces it. Report figures remain. |
| Deleted workspace content | Models, files, datasets, and API keys removed immediately. Workspace name, billing history, and the deletion Activity entry remain. |
These are live-Service deletion periods. Deleted storage versions can remain for up to 30 days; protected backups follow the periods in the Privacy Policy and DPA. Deleting an upload does not delete derived datasets, resized pictures, thumbnails, representations, or trained models.
Read the full retention policy →If you connect your own AI platform and use Your AI Model, or train a model in your own Amazon Bedrock, StayCharted sends the rows that model is trained on, tested on and asked about to that platform, on your instruction. The provider is engaged by you under your own agreement and is not a StayCharted subprocessor. Keys are stored encrypted and never shown again; Amazon Bedrock connects through a role in your AWS account that only AI Model Trainer can use. A model trained in your Bedrock stays in your account, and its training file is deleted from your bucket when training ends.
StayCharted reads only the tables your Owner lists, and writes only one results table, in a schema you name; it never changes a table of yours. The grant SQL it gives you allows exactly that, so your warehouse enforces it too. Snowflake signs in as a service user with a key pair StayCharted makes: the private key is stored encrypted and never shown, and you set only the public half. Databricks signs in as a service principal whose OAuth secret is stored encrypted and never shown again. StayCharted calls only Snowflake's and Databricks' own addresses, and every value in a query is passed as a parameter, never pasted into the SQL. Rows read to train or fill are kept and deleted as an uploaded file is.
The app signs in through AMT's own sign-in page, with your usual password, Google, Microsoft or Apple sign-in and two-factor; it never sees your password. An access code lets a phone use one model and nothing else, with limits on days, phones and answers a day that a Builder sets. A phone presents its code once and then uses its own token, kept in the phone's secure storage and stored on our side only as a hash. Builders can pause or revoke a code or remove a phone at any time, and the Owner can switch codes off for the workspace. Making, pausing and revoking codes, and every phone that joins or is refused, are in the Activity log. Photos are used to answer and aren't kept, unless an unsure answer goes to the Review Queue.
Customer Content hosted by StayCharted is stored and processed in the United States. Rows sent to your connected AI platform are processed where that platform runs, under your agreement with its provider.
The Privacy Policy describes TLS in transit, encryption at rest for the database and content storage, and organization and workspace isolation.
The DPA sets out processing obligations, security measures, and subprocessors. The legal documents govern the details summarized here.
| Document | What it covers | Read or request |
|---|---|---|
| Data Processing Addendum | Processing terms, security measures, subprocessors, EU SCCs and UK Addendum. | Read the DPA |
| Privacy Policy | What we collect and why, retention and your rights. | Read the Privacy Policy |
| Subprocessors | AWS hosting and Resend notification email, with advance notice of changes as set out in the DPA. | View subprocessors |
| Self-hosted deployment | For organizations that need AMT in their own AWS account and Region. On request, as part of Enterprise. | Tell us what you need |
| A completed questionnaire | Security and data protection questionnaires; standard answers follow the Cloud Security Alliance questionnaire (CAIQ). | Request a questionnaire |
StayCharted does not hold SOC 2 or ISO 27001 certification. AWS’s certifications cover AWS’s infrastructure, not our application, so we do not claim them as ours.
StayCharted
3801 N Capital of TX Hwy Ste E-240
Austin, TX 78746, United States
If you think you have found a vulnerability, email security@staycharted.com with the details. We will confirm receipt, keep you updated, and will not take action against good-faith research that avoids other customers’ data and does not disrupt the service.
StayCharted hosts Customer Content on AWS in the United States. If you connect your own AI platform, the rows you ask it to process go to that platform under your agreement with its provider. The Privacy Policy and DPA describe these arrangements.
Only your own. Your examples train the models in your workspace, and nobody else’s. The shared pretrained AI models do not learn from your uploads. If you connect your own AI assistant or AI platform, it receives what you send it under your own agreement with its provider.
Staff cannot sign in as you or act inside your account; the product has no mechanism for that. Administrative access requires multi-factor authentication (MFA) and is controlled by role. Authorized infrastructure access is separate, and administrative actions are logged.
Yes. Email support@staycharted.com. Our standard answers follow the Cloud Security Alliance questionnaire (CAIQ), and our DPA sets out the security measures we commit to.
Retention depends on the data. Original uploads are kept for 90 days unless deleted sooner; filled files are removed 24 hours after first download or after 7 days without a download. Derived datasets and models have separate retention. Deleted storage versions may remain for up to 30 days; see the retention table and Privacy Policy for the full rules.
StayCharted does not hold SOC 2 or ISO 27001 certification. See the security and compliance documents above for our documented measures and questionnaire contact.
Free includes text Classifier Models and one AI Classifier Model. Essentials adds API access, the Review Queue and starting without answers; Business adds Dedicated AI Models. Vision is an optional add-on on paid plans.