Agentic AI you can put in front of counsel.
Brandeis runs a staff of AI agents on public records — where “the model decided” is never an acceptable answer. So the guardrails aren't promises. They're built into the software and tested on every change.
Agents propose. A named human disposes.
Every AI output is a reviewable draft — an accept, edit, or dismiss card. Nothing legally significant leaves the building without a named person deciding it should.
Action tiers, enforced in code
Anything legally significant or requester-facing is always a human's to do. No setting, flag, or approval can hand it to an agent — the software has no path for it.
Approvals are per step
A staffer approves one step, not a whole run. The next sensitive step stops and waits for its own approval.
Every agent runs on a budget
There is a hard cap on what a run can do before it must stop and report. A runaway agent is a logged pause, not an incident.
Provenance on every AI output
Every AI action records what it ran on and what came of it. When counsel asks why the AI suggested something, the answer is a log entry, not a shrug.
Nothing becomes public by machine
An agent may propose that a record is public; only a named human can make it so. No automated path exists in the other direction.
No AI required at all
Switch the AI off and the full records desk still runs. The agents are leverage, never a dependency.
Two steps for every staffer. No exceptions to configure.
A stolen password alone opens nothing.
A code on every staff sign-in
A password, then a one-time code emailed to the staffer. It's required for every office, and no setting turns it off.
Sessions that end
Thirty idle minutes or twelve hours, whichever comes first. A password reset signs that person out everywhere at once.
Every sign-in on the record
Failed passwords, codes sent, sign-ins and sign-outs land in the same permanent log as the work, readable by your admins.
Pages also ship with standard browser protections: HTTPS-only, no framing by other sites, and no camera, microphone or location access. Next on this list: authenticator apps and passkeys, single sign-on through Microsoft Entra or Google, and promoting the browser content policy from report-only to enforced.
Built to be audited.
Append-only audit log
Events can only be added, never changed or removed — the software has no way to do either. People and AI land in the same record.
True redaction
A finalized redaction is part of the page itself, not a box a reader can lift off. Scans are rebuilt from the burned pixels, with no hidden text layer. Every file is checked for leftovers before release.
Releases anyone can verify
Every release is fingerprinted when it's approved. A court or newsroom can check a copy in the browser — the file never leaves their machine.
One platform, every agency its own house.
Tenant isolation, proven by tests
One agency can never see another's records, and the tests we run on every change try to cross that line on purpose.
Public means public, everywhere
The portal, the answer box, and the door AI assistants use all reach exactly the same thing: records that are already public.
Self-contained by design
No outside service is required to run it. Host it yourself and your records never leave your building.
The product itself carries no third-party trackers at all — no portal, staff, or console page measures anyone. This marketing site does use Google Analytics, and it honors Do Not Track: set it and nothing loads. Our own measurement stores no visitor ids and no IP addresses.
Bring the hard ones.
Security questions, disclosure reports, or a walkthrough of the guardrails on the live product: [email protected].
Or see the audit trail yourself on the live demo.