Security
Your clients' financials are not a place to be vague.
What we do today to protect a firm's data, in plain words, with the limits stated. Where a page describes something we have not done, it says so.
Where it lives
Hosted in Sydney
The database, sign-in and file storage run in Australia, and so do the application servers.
Database, authentication and file storage with Supabase in Sydney (ap-southeast-2).
Application servers with Vercel in Sydney (the syd1 region). Pages pass through Vercel's global edge network on the way to you.
Some supporting services, such as error monitoring, product analytics and background jobs, are based in the United States. They are listed with their purpose on the sub-processors table.
AI features are processed by Anthropic in the United States. See AI, and what it sees.
In transit and at rest
Encryption
Data is encrypted on the way in and while it is stored.
All connections use TLS 1.2 or later.
The database and file storage are encrypted at rest.
Credentials for connected services are encrypted with AES-256-GCM and cannot be read back through the application.
Roles and sign-in
Who can see what
Each firm is its own workspace, and inside it people see what their role allows.
- Row-level security on every table that holds a firm's data, so one firm cannot read another's.
- Roles per workspace (owner, admin, member and viewer), with visibility groups and folder permissions on top.
- Sign-in by email link or password, and two-step sign-in with an authenticator app for anyone who turns it on in Settings, Security. It is required for our own platform operators, and re-checked when a workspace's ownership is transferred.
- Our staff have no standing access to a workspace. A support session is read-only, limited to 60 minutes, opened with a recorded reason, and visible to the workspace's owners and admins, who can end it at any time. Financial statements, uploaded documents and data-room files are outside its scope.

Demo workspace, invented data
The record
Audit log and backups
What happened is recorded, and the data can be restored.
- An append-only audit log: the database refuses to change or delete a row once written. Decisions and overrides are recorded with their reason, and each engagement has an export PDF.
- Data-room opens and downloads are recorded in the documents access log.
- The database is backed up by our database provider, and there is a written backup and restore runbook.
- A workspace's owner can export its data and files, and delete it, with a 30-day grace period.

Demo workspace, invented data
Guardrails
AI, and what it sees
AI features are processed by Anthropic in the United States, with controls on what is sent and whether it is used at all.
- People's names, email addresses and phone numbers are replaced with tokens before most AI requests, and put back only for the people allowed to read the result. Business names are kept, because drafting needs them.
- The honest exceptions: statement PDFs and images sent to the model as files cannot be tokenised, so treat an upload as sent to the United States route; a person who is not in your contacts and is not an email or phone number cannot be recognised; and the phone receptionist needs a caller's name and number to do its job.
- AI can be switched off for a whole workspace, and separately for a single deal, for example on an expert or court engagement where you may need to say that no AI drafted anything. Every AI-drafted IM section records who drafted, edited and accepted it.
The small print, stated
Who else touches the data, and how to tell us.
Every service is declared.
Every service that receives personal data on our behalf is listed with its purpose and region in the sub-processors table (section 6) of our data processing addendum, generated from the same declaration the application uses. A new integration cannot be added without being declared there.
See the sub-processorsPlanned, not claimed.
SOC 2 / ISO 27001: planned. We have not announced a date, and we do not claim it today.
Found something? Tell us.
Email team@xlev.ai with what you found and how to reproduce it. Please do not access other people's data or disrupt the service while you test. If a breach affects your data, the data processing addendum commits us to tell you within 72 hours of confirming it.
Questions
Security: what people ask.
Where is my data stored?
In Sydney: the database, file storage and sign-in with Supabase, and the application servers with Vercel. A few supporting services are based in the United States, and AI features are processed by Anthropic in the United States. The sub-processors table in our data processing addendum lists each with its purpose and region.
Is two-step sign-in required?
Anyone can turn on two-step sign-in with an authenticator app in Settings, Security. It is required for our own platform operators. It is not yet required of each workspace's owners and admins.
Does the AI see my clients' names?
Before most AI requests, people's names, email addresses and phone numbers are replaced with tokens. There are honest exceptions, such as statement PDFs sent to the model as files, listed on this page. AI can be switched off for a workspace or for a single deal.
Are you certified?
Independent certification (SOC 2 or ISO 27001) is planned. We have not announced a date, and we do not claim it today.
How do I report a security issue?
Email team@xlev.ai with what you found and how to reproduce it. Please do not access other people's data or disrupt the service while you test.