Nothing is required
Every agent on eCourt AI works inside a control plane your agency owns: permissions you set, approvals you require, autonomy you dial, and a record of everything.
Governance
Your policies, not ours. You define what each agent may do, which workflows it touches, and which staff supervise it. Role-based permissions cover agents the same way they cover people.
Observability
Watch your agents work in real time. Every agent has a manager view showing what it is doing, what it has queued, and what it is waiting on, and the manager is you.
Traceability
Everything on the record: every action attributed to the agent that took it and the human who approved it, timestamped, reviewable, exportable. Courts run on records. So do our agents.
Controls and options
Nothing is required. Every capability is opt-in, every automation can be turned off, and every setting is yours to change without a support ticket.
Autonomy is a dial
Set per agent and per workflow, changeable anytime. Agents earn autonomy the way staff do: by doing the work well, visibly, over time.
Suggest
the agent recommends, your staff act
Draft for approval
the agent prepares, a human signs
default
Act with review
the agent acts, your staff audit
Act on its own
earned, scoped, always on the record
A word about the fourth level. Act on its own is never a default and never required. An agency that uses it turns it on per workflow, scopes it to routine and reversible work, and can turn it off with one setting. Everything it covers stays attributed, timestamped, and reviewable, so the record reads the same whether a person or an agent did the work, and a person is always accountable for it.
When an agent is wrong
Agents make mistakes, and a system built on the record is built to catch them. In a court, a wrong answer is not a footnote: it can cost a person a hearing date. So the correction path is part of the control plane, not a support ticket.
Caught on the record
Every answer an agent gives cites the record it came from, so a wrong one is visible, not silent. Staff flag it in the same interface they approve in, and the manager view shows anything questioned or corrected.
Corrected once, everywhere
A correction is another update: entered once, approved by a person, and landed everywhere the original went. The notice, the calendar, the portal answer, and the person affected, reached in their language.
Accountable to a person
The error and the correction both stay on the audit trail, attributed and timestamped. And the dial answers risk: workflows where a wrong answer costs most can stay at suggest or draft for approval, where nothing reaches the public without staff review.
Your jurisdiction’s rules. Your local policies.
Governance and compliance are configured to your jurisdiction: the statutes above you, the policies your agency sets, local court rules, and whatever AI policy your jurisdiction has adopted. One system runs an entire jurisdiction, with every local court on its own terms, which is why jurisdiction-wide deployments work without flattening local practice.
Humans in the loop, by policy
Agents draft, prepare, flag, answer, and triage. They never decide. Judgment, discretion, and authority stay with your staff, and the approval moment is built into the interface, not bolted on. Your team doesn’t work for the software; they staff agents that work for them.
See the control plane in your context
Start with one agent, connect what you have, and keep your staff in control from the first day. Add the next role when the results make the case.
Not sure where to start?Get an AI readiness plan