Agents you can scope, supervise and audit.
Not a chat window with a personality — a scoped role with declared skills, an autonomy ceiling enforced at runtime, and a named supervising lawyer.
What stays true no matter which model runs.
Tier is scope, not judgment
Tier describes how large a task an agent may be given. No output is relied on without a qualified person reviewing it.
A named supervisor
An agent cannot be assigned to a matter until a practitioner is bound to it. The binding travels with every action.
Memory does not cross matters
Configuration is reusable; memory is not. It is isolated at provision and sealed at close.
Autonomy capped in the runtime
Draft-with-approval is a constraint the system enforces, not an instruction the agent can be argued out of.
Supervision
The gate is where the agent stops.
An agent drafts; a partner releases. Four-eyes review on AI work is a state of the matter, and the reviewer, the version and the reason are all in the record before anything reaches a client.
- Held pending review, by default
- Accept, modify or reject — each recorded
- The gate cannot be skipped under deadline
Billed as what it is
Agent work never becomes an attorney hour.
Every task an agent performs is a separate line item at its own rate, attributed to the agent that ran it. The platform will not aggregate the two.
- Attributed to the named agent and matter
- Recorded as a task, never as hours
- Disclosable to the client as a line
Qualification
An agent is evaluated before it is offered for a skill.
Skills are explicit — drafting, arbitration, pleadings, cite-check — and each is scored against a task set before the agent can be assigned to it. We publish the method, and you can re-run it on your own matters. Out-of-scope requests are refused and logged, not attempted.
Governance metrics
Your numbers about your work.
Four-eyes pass rate, citation verification and escalation rate, computed from your signed audit trail rather than from a vendor benchmark.
Local compute
Concurrency is a capacity question, not a bill.
Agents run on hardware you own and are licensed per firm — the capability is yours, and so is the compute.
A clause is the usual answer. Ours is a clause and an architecture.
On every tier. The matter boundary, the approval gate and the signed record are enforced in the runtime — hosted, in your own infrastructure, or on Bernio Pro.
- ABA Model Rules 5.1 & 5.3US
Managerial lawyers must make reasonable efforts to ensure subordinate and nonlawyer assistance is compatible with professional obligations.
An agent cannot act on a matter without a bound supervisor, and every output stops at a gate that supervisor owns.
- ABA Formal Opinion 512US
Lawyers must maintain competence in the tool, protect confidentiality, and independently verify generative output.
Verification is a state the work passes through, and the verification itself is recorded against the version.
- GDPR Art. 25EU
Data protection by design and by default: processing must be limited to what each purpose requires.
Per-matter memory isolation and sealing at close are the default state of an agent, not a setting a firm must find.
Put an agent on a matter you already closed.
The most useful demo is a matter whose outcome you know. We staff it, supervise it, and show you the record.
Development status
Bernio is in active development with design partners — firms running real matters through the platform and shaping what is built next — and the first commercial contracts are being signed on founding terms. The capabilities described on this page are built and can be demonstrated on a matter type your team handles. Bernio Pro — the on-premise appliance — is an engineering prototype and is not yet commercially available, so any capability on this site that depends on it is marked as such. Descriptions of the product are not warranties; the binding terms for any deployment are those set out in the applicable written agreement.