One agent per case. It knows the matter and nothing else.
Each case has its own AI assistant. It works from the documents you give it, keeps its findings in a review queue and hands decisions back to you.
- Clause 7.2 requires 14 days’ written noticeProposed
↳ Supply agreement v2.pdf · p.6
- Delivery date agreed as 18 SepConfirmed
↳ Email thread · 02 Sep · confirmed by reviewer
- Letter refers to immediate terminationNeeds source check
↳ Letter before claim.pdf · p.1
// A proposed finding never becomes part of a draft or report until a person confirms it.
What a Case Agent can do
- AGT-01Core
Dedicated agent per case
Each case has its own assistant with access to that case only. No cross-case memory.
- AGT-02Core
User-requested private document analysis
Your documents are analysed only when you ask, inside the case, with the action logged.
- AGT-03Core
Human-reviewed recommendations
Findings, drafts and proposed tasks are recommendations until a person approves them.
- AGT-04Core
Proposed versus confirmed findings
Findings carry a state: proposed by the agent, or confirmed by a reviewer. The timeline shows which.
- AGT-05Planned
Scheduled checks on a case
The agent reviews new material in a case on a schedule and reports what changed. Disabled by policy until validated.
- AGT-06Planned
Autonomous approved-source legal research
Agents monitor authorised public legal sources and add verified, cited material to a knowledge library.
- Core
- part of the workspace design shipping first
- Early access
- available to early-access workspaces, with limits
- In development
- being integrated and tested; not yet usable
- Planned
- designed; released only after validation
Three modes, one rule: people decide
The distinction that matters: autonomous research only ever touches approved public sources. Your private documents are processed only when you ask.
- On requestCore
Analysis of your private case documents
Your documents are analysed only when you ask, inside the case they belong to, with the action logged in the timeline.
- ScheduledPlanned
Checks on a case you own
Planned: the agent reviews new material in a case on a schedule and reports what changed. Scheduled runs are disabled by policy until validated; today agents run only on an explicit request.
- AutonomousPlanned
Research of approved public sources
Planned: research agents read authorised public legal sources on a schedule and add verified, cited material to a knowledge library.
- NeverBy design
Sending correspondence or deciding for you
Agents do not send letters or emails, file anything, or make professional judgements. They prepare; people decide.
Every agent action is attributed
Who or what did it, on which document, at what time, with which sources. Agent actions appear in the case timeline next to human ones and are included in exports.
Ready to look at your matter more clearly?
Plans start at £29 per month in the UK, billed per workspace. Choose a plan now; checkout opens with the application.