Checked by a different model
Before an agent finalises anything, a second model from a different family re-derives each finding from only the passages it cited. A model checking its own work shares its blind spots; a decorrelated one does not.
Matter Agents
HeadlineHand off the long jobs. Keep the judgement.
Supervised agents plan, work, and have their findings re-derived by a different model family — so the only things that reach you are the decisions that need a lawyer.
01What it is
Matter Agents are supervised, long-horizon agents that take on multi-step jobs — diligence passes, document triage, research sweeps, standing monitors — and bring back decisions, not surprises. They plan, work, are independently checked, and promote cited work into the matter.
02In detail
Escalates when it is unsure
A human is pulled in on measured grounding failure or disagreement, not after an arbitrary number of steps — and the agent gets exactly one bounded attempt to correct itself before it stops and asks.
Configured, versioned, dry-run
Describe the job in a guided interview; the platform selects skills, sets the cadence and the guardrails, freezes the configuration as a version, and runs a dry run you review before anything is deployed.
Three scopes, one engine
The same agent runs against a matter, your personal workbench, or a firm-wide knowledge base — on demand or on a schedule as a standing monitor. Scope and cadence are settings, not different products.
03In practice
Provenance
Supervised, by default.
The loop pauses for a human at the gate that matters — and it is a second, independent model that decides which gate that is.
A supervised Matter Agent cycles through Plan, Execute, Reflect and Promote, pausing at a human approval gate before promoting cited findings into the Matter Wiki.
Matter Wiki
124
cited entries indexed this cycle
Each promotion writes a finding — with its source anchor — into the matter's living memory.
04Why it matters
How it fits the matter
Supervision is the point, and it stays the public posture: the agent does the legwork and proves its own grounding, you make the calls that carry judgement, and the matter keeps everything it learns.
Works with
Skills & Content PacksPractice-area packs, firm-authored skills, and playbooks the agent actually executes.ExploreFirm MemoryCross-matter institutional knowledge, wall-governed at read time and exportable in an open format.ExploreWorkflowsReusable, governed multi-step automations over your sources and tools.Explore