Built for privileged material from the first line of code.
Client confidences are the starting assumption. Here are the controls, described plainly, and the commitments that govern how AI is used.
Each firm is sealed off in the database itself.
Every record that belongs to a firm, from cases and documents to findings and audit entries, carries the firm's identifier. The database enforces row-level security on every query, so a request made in one firm's context can't read or write another firm's rows, even if application code has a mistake.
The application connects with a database role that can't bypass these rules and doesn't own the tables. Automated tests try to reach another firm's data through every endpoint and directly in the database, and fail the build if they succeed.
Controls.
Private storage
Documents are stored in a private bucket, encrypted in transit and at rest. They open only through links that expire within minutes, issued after a permission check. Every download is recorded.
Uploads are checked
Files are identified by their contents, not their extension, then fingerprinted and size-checked before processing. Duplicates are skipped, and rejected files are deleted rather than kept.
Accounts and sessions
Passwords are hashed with Argon2id. Sessions are held on the server and can be revoked from any device. Sign-in is rate-limited, and every state-changing request is protected against cross-site forgery.
Roles and case access
Owners, administrators, attorneys and paralegals have defined permissions. By default, attorneys and paralegals see only the cases they're assigned to. Nobody can grant a role more senior than their own.
Tamper-evident audit trail
Uploads, downloads, edits, approvals and deletions are written to an append-only log. Each entry is chained to the one before it, so any alteration is detectable. Administrators can review and export it.
Retention and deletion
Deleting a case permanently removes its stored files and data, after the person deleting it confirms the case number. The audit record of the deletion is kept.
How AI is used, and how it isn’t.
- 01
No training on your files
Your documents are never used to train AI models, ours or anyone else's.
- 02
Zero-retention terms before client data
Client files are sent to AI model providers only under contractual zero-data-retention and no-training terms. Prompts and outputs aren't stored in logs.
- 03
Documents can't give instructions
Text inside an uploaded file is treated as evidence only. It can't change instructions, permissions or the legal framework, and analysis steps have no ability to act outside their task.
- 04
Every output is validated
Model responses must match a strict structure, quotes must exist on the cited page, and citations must come from the approved framework. Anything else is rejected.
What we don’t claim.
We don't claim certifications we haven't earned. There are no compliance badges on this site because no independent audit has been completed yet. When one is, it will be listed here with its scope and date.
Firms in the pilot program receive our security documentation, a list of the third parties that process data, and direct answers to their questionnaires.
Ask for the security documentationBring your security questions to the first call.
We'll walk your team through the controls in detail.