Where do tests run?
On your machines, inside your network.
Whose AI account?
Yours. Set up by us, billed to you.
Where do credentials sit?
AES-256-GCM. Entered by reference.
Who makes the ship call?
A person. The agent brings evidence.
Which surfaces?
Web, mobile, and desktop.
What is kept per step?
Action, notes, screenshot, outcome.
Does the model keep prompts?
No. Zero data retention.
Who can see a project?
Its members, at the role you assign.
A payment in production?
Refused. Every time.
Is our data used for training?
Never. Not by us, not by providers.
What waits for a person?
Test plans, deletes, and submits.
Where is project data stored?
Your cloud, or a dedicated environment.
What does the model see?
Masked page content. No card or Aadhaar numbers.
Your data stays inyournetwork.
Studios Pro runs tests on your team's machines and keeps project data in an environment you choose. AI runs on your own account: masked content goes out, credentials never do.
- CredentialsNever sent to a modelPasswords and secrets are encrypted with AES-256-GCM and entered by reference. The model sees the name, never the value.
- AI modelsNo training. Zero retention.Every request runs on your own AI account, routed only to approved providers that keep nothing once the response returns.
Only the tools you permit.
The agent reaches the inbox, number, channel, server and model you switch on for this project, and nothing else. A connection that is off is not in the walk.
- Email inboxAdded
A test address the agent owns, so it can read the sign-up and reset emails your app sends.
- SMS numberNot added
A dedicated number for one-time codes and alerts, with a spend cap.
- WhatsApp numberNot added
Template messages to an allowlist you set.
Slack channelAddedWhere run reports land.
- Teams channelNot added
Where run reports land.
- MCP serverAdded
Your own tools, added to this project only.
- Model provider you approveAdded
Approved providers, through an OpenRouter account in your name.
No training. Zero data retention.
- VaultAlways on
Credentials stay here, apart from product memory.
AES-256-GCM. The model sees a name, never the value.
What the agent never touches
These hold whether a connection is on or off.
01Credentials
- Path
- vault://checkout/staging
- Secret
- ••••••••••••
- Held out
- Not in the prompt · not in memory
- Redacted
- Tool output, before the model
The whole program on one screen.
Who is working, what the runs are spending against the budget you set, and which packages are open. The people who sign off read the same picture the team works from.
- Localization manager
- Linguist
- Reviewer
- Engineer
In the workspace
Localization manager · Linguist · Reviewer · Engineer
- Your key
- Budget · 62% used
Token spend
Last 14 days
5.5M
tokens · 76% walks · 24% localization
Readiness
This week against last
82%
Journeys walked
Last 12 months
A sample workspace. Not a live account.
Packages in use
Formats we acceptWith the project
What the walk is allowed to open.
- PNGcheckout.png4.1 MB
- PNGrtl-ar.png3.8 MB
- TXTrun-note.txt2 KB
Where a test actually runs.
Tests run on your team's machines, inside your network. The server runs the agent and keeps project data in your Studios Pro environment. Only the AI step leaves, masked, on your own account.
What a reviewer can check.
Three areas your InfoSec team will ask about: who can do what, how data is protected, and how AI use and spend stay under your control.
- In the product
- You configure
- On request
Access & governance
Your administrators decide who sees each project and what the agent may do. Every approval and admin change is recorded.
5
built-in roles, plus custom
- In the product
- 4
- You configure
- 1
- On request
- 1
- Tester
- Lead
- Manager
- Owner
- Super admin
- TerminalOff
- System settingsOff
- Coding agentsOff
Granted by a lead or administrator.
Invitations, role changes, approvals, and other admin events are recorded.
Role-based access
In the productTester, Lead, Manager, Owner, and Super admin, plus custom roles. Users see only the projects they are added to.
Sign-in
In the productMulti-factor authentication is built in. Passwords are stored as salted hashes. Removing a user revokes their sessions immediately.
Single sign-on
On requestConnects to your identity provider, so access follows your joiner, mover, and leaver process.
Approval gates
In the productTest plans stop for approval. Deletes, submits, and other irreversible actions need explicit confirmation.
Sensitive capabilities
You configureTerminals, system settings, and coding-agent integrations stay off until a lead or administrator grants them. Any agent tool can be disabled per project.
Audit log
In the productInvitations, role changes, approvals, and other admin events, exportable by administrators.
What the review covers.
Five questions from a review. The rest are in the help center.
Only the AI step leaves your Studios Pro environment: your instructions, masked page content, screenshots of the application under test, and tool results after secret redaction. It goes through an AI gateway account in your name to approved providers, under zero data retention. Passwords, secret values, API keys, raw vault contents, and your AI gateway key are never sent. Your data is never used to train, fine-tune, or evaluate any model.
In your Studios Pro environment: your own cloud account, or a dedicated managed environment in the region you choose. Run history, screenshots, traces, recordings, and audit events are kept for two years by default, and every period can be changed organization-wide or per project. Project memory stays until the project is deleted, and saved browser login sessions expire after 7 days. Deletion pauses under a legal hold on written request.
It is not shared SaaS. KaDeep deploys the Studios Pro server in your own cloud account, which we recommend, or runs a dedicated managed environment for you in any region you choose. Tests always run on your team's machines, inside your network. If you need a deployment in your own data center rather than a cloud account, raise it in the walkthrough and we confirm what fits.
Both, with people in charge. Routine test steps run within the scope you approve. Test plans need approval, irreversible actions need confirmation, and payment actions in production are refused. Users can pause the agent or take over at any point, and the agent hands control back when it is blocked or unsure. Page content is treated as data, never as instructions, so injected content cannot bypass these gates.