What your IT team needs to know before a pilot.
Where tests run, what has to reach what, what is installed, what we store, how the AI is used, and what we haven't confirmed yet. Send this page to your desktop, network and security teams; a pilot starts when they are satisfied.
1. Where tests run, and what must be reachable
This is the question that decides whether a pilot can start, so it comes first.
| What | Where the browser runs | Network |
|---|---|---|
| Record, pick an element, Watch a run | The tester’s own browser, opened by the agent on their computer | Inside your network. Works behind your VPN. |
| Run from the app (not watched), suite runs, schedules, atx run --suite | A headless browser on the AutoTestX server, AWS Canada (Central) | The server must reach your test environment’s URL. |
| atx run <test.json> in your own CI runner | The machine that runs the command, with no AutoTestX server involved | Inside your network. Results stay on that machine as JSON. |
Unattended runs need your test environment to be reachable from the AutoTestX server.The usual options:
- Allow the AutoTestX server’s fixed outbound IP address through to your Maximo test environment (we give you the address during the pre-check).
- On Business, a dedicated instance of your own in AWS Canada or the US, with its own fixed address.
- Run tests with
atx runin a CI runner inside your network. Results are JSON on that machine, not in the app.
If none of these is acceptable, recording and watched runs still work behind your VPN, but scheduled and suite runs will not. Tell us early: it changes whether a pilot makes sense.
From testers’ computers: HTTPS to app.autotestx.com on port 443, for the browser and for the agent’s one WebSocket connection. No inbound ports.
2. The agent on testers’ computers
- Needed for recording, picking an element and watching a run. Not for anything else.
- Windows: a download (about 80 MB) that installs for the current user only, with no administrator rights, starts at sign-in, and is removed from Settings → Apps.
- macOS: installed with one Terminal command, for the current user only, removed with
--uninstall. - Signing: neither installer is code-signed yet, and the Mac app is not notarized by Apple. SmartScreen or your endpoint protection may warn about or block it; plan for an allow rule.
- Browser: opens Edge (Windows) or Chrome (Mac) with a temporary, separate profile, so it never touches the user’s own browser data.
- Pairing: each computer is approved by the user, signed in, in their browser, and can be disconnected at any time under Profile → Your computers.
- Logs:
%LOCALAPPDATA%\AutoTestX\agent.logor~/Library/Logs/AutoTestX/agent.log.
The agent in the documentation →
3. What is stored, and where
On one server in AWS Canada (Central), as files on its disk, backed up by daily server snapshots. Business plans can have a dedicated instance in Canada or the US. All traffic uses HTTPS.
| Data | Stored |
|---|---|
| Test definitions, components, version history | Yes |
| Environment settings, including secret values for your test system | Yes. Kept server-side, never returned to the browser; not encrypted at the application level today |
| Test data you upload or attach | Yes |
| Run results, step screenshots, logs | Yes, until deleted. No automatic retention yet |
| Video and trace of failed or flaky runs | Yes, deleted after 14 days |
| Names, emails, roles; password hashes (scrypt) | Yes |
| Your organization’s own Anthropic key, if saved | Yes, encrypted, shown only as its last four characters |
AutoTestX is for test environments. Use test data, not production data or real people’s personal information.
4. AI
Optional. An organization admin can turn AI off for the whole organization, and AI healing is off until a project turns it on. No run acts on an AI answer.
- Which features: the AI step writer, AI healing suggestions and AI failure diagnosis.
- Provider: Anthropic’s API (United States), on the AutoTestX key or your organization’s own Anthropic key. On Enterprise, only your own key.
- What is sent: the text of a test you ask it to draft; the names and attributes of page elements (role, label, id, heading) for healing; the failed step’s details for diagnosis.
- Never sent: screenshots, recordings or secret values.
5. Screenshots, video and what is redacted
Be clear on this one: redaction applies to text, not to pictures.
- Text: secret values are replaced with
***REDACTED***in every log line, step detail, error message and variable in a report, and masked inside traces. - Screenshots are not masked. A step screenshot shows whatever was on the screen: record details, names, amounts, and any value typed into a field that displays it. Password fields show dots, as in the browser.
- When screenshots are taken: after every step for runs from the app, on failure for
atx rununless--screenshotsis given, and always for Screenshot steps. - Video and trace are kept only for failed or flaky runs, for 14 days. If a step types a sensitive value into a field that shows it, the run keeps no video or trace at all.
The remaining risk is data on screen. Keep it to test data, and type credentials only into password fields.
6. Accounts and access
- Email and password accounts. Passwords stored as scrypt hashes; sessions in HttpOnly, SameSite=Strict, Secure cookies that end after 12 hours idle, 7 days at most.
- Each customer is a separate organization. Every record is checked against its own project; refused cross-tenant requests are logged.
- Five project roles (Admin, Test Manager, Test Developer, Tester, Viewer) and organization admins. Only roles with the permission can add custom code steps.
- Self-healing decisions are kept in an append-only, hash-chained audit trail.
7. Export and deletion
- Tests are JSON in a documented format, which you can copy out at any time.
- On request we export or delete your organization’s data within 30 days. Deletion is done by us, by hand, today; there is no self-serve delete or automatic retention yet.
- Data still in server snapshots is overwritten as the snapshots roll over.
8. Not confirmed yet
What we can’t yet state, so you don’t have to discover it:
| Question | Today |
|---|---|
| Encryption at rest of the server disk | Not confirmed yet. We will confirm it in writing for your review. |
| Retention period of the daily server snapshots | Not confirmed yet. |
| Code signing of the Windows and Mac agents | Not signed yet. Your endpoint protection may need an allow rule. |
| Penetration test, SOC 2, ISO 27001 | None yet. |
| Single sign-on and multi-factor authentication | Not built yet. Email and password accounts. |
| Running hosted suites through a machine inside your network | Not available. Use the reachability options below, or atx run in your own CI. |
A security questionnaire, or a question not answered here? Emailinfo@autotestx.com or use the contact form.
Questions from your IT team? We'll answer them before a pilot.
In 30 minutes: a live demo of AI drafting, healing and explaining tests on a representative flow, and what your team would need for a pilot.