The Predictive Legal Agent. Drafts a senior associate would.

Generic LLMs can't do this.

What it actually does.

Real work the agent does end-to-end. Every step is auditable, every consent gate logged, every escalation routed to a named human.

1. 01

   NDA and vendor-paper triage

   First-pass redline on inbound NDAs and standard vendor agreements, routed by risk tier.

   • Intake form classifies counterparty, deal value, and template (yours or theirs).
   • Agent extracts the 40 to 60 deal points the team cares about, scored against your playbook.
   • Green-tier NDAs return a clean redline and signature-ready draft in under a minute.
   • Yellow-tier surfaces non-standard clauses with a cited rationale from your playbook.
   • Red-tier (indemnity carve-outs, unusual IP clauses, data terms) escalates to a named lawyer with a written summary.
   Confidentiality gateDocuments processed in your tenancy, encrypted at rest, never used for cross-client training. Privilege markers attach automatically to draft outputs.

   Human in the loopCounsel signs every NDA before send. The agent only proposes.
2. 02

   Policy compliance check

   Tests draft commercial terms, marketing copy, and product changes against your live policy library.

   • Business stakeholder submits the artefact through the intake form.
   • Agent retrieves applicable internal policy paragraphs and external regulatory text.
   • Each finding is returned with a paragraph-level citation and a confidence score.
   • Conflicting policies are surfaced rather than silently resolved.
   • Counsel approves, edits, or refuses.
   Confidentiality gatePolicy library indexed once, queried in tenant. No prompt or response leaves the customer's region.

   Human in the loopCounsel decides which findings are material and how to advise the business.
3. 03

   Regulatory diff watch

   Tracks regulator publications and re-tests existing contracts and policies when wording changes.

   • Subscribes to SEC, EU AI Act, DSA, GDPR, India DPDP, and sector regulators by jurisdiction.
   • Detects substantive changes (not formatting) against the previous version.
   • Maps each change to affected contracts, policies, and prior advice.
   • Generates a counsel-facing memo: what changed, what it touches, suggested action.
   • Adds the change to a tracked register with owner and due date.
   Confidentiality gatePublic sources stay public. Mapping happens against your private corpus inside your tenancy.

   Human in the loopCounsel decides which change is material and which letters or contract amendments go out.
4. 04

   Contract redline against playbook

   Counterparty paper compared clause-by-clause to your house playbook, with cited fallback positions.

   • Agent ingests counterparty draft and your playbook for that contract type.
   • Clause-level embeddings find your nearest sanctioned position.
   • Drafts a redline with preferred, fallback, and walk-away positions inline.
   • Each suggestion cites the playbook paragraph and any prior matter where the position held.
   • Surfaces the three clauses most likely to drive negotiation cycles.
   Confidentiality gatePlaybook and counterparty paper stay in your data perimeter. No model provider sees the file body.

   Human in the loopCounsel owns the redline that goes back. The agent's role ends at the first-pass draft.
5. 05

   Matter intake and early case assessment

   Triages new disputes against the history of how similar matters resolved.

   • Intake captures parties, claim type, jurisdiction, exposure estimate.
   • Agent retrieves comparable closed matters from your history.
   • Returns base rates: settled vs litigated, average duration, ranges of cost and recovery.
   • Flags clause patterns or counterparty behaviour previously linked to escalation.
   • Drafts a one-page early case assessment with citations to the prior matter files.
   Confidentiality gateMatter files carry privilege markers. Outputs are tagged as work product where the test in Warner v. Gilbarco is met.

   Human in the loopLead counsel decides strategy and budget. The agent's base rates are inputs, not decisions.
6. 06

   RFP and security questionnaire response

   Drafts answers to inbound legal, security, and AI-governance questionnaires from prior approved responses.

   • Parses the inbound questionnaire into individual questions.
   • Retrieves the most recent approved answer for each question, with the date and approver.
   • Marks any question whose answer is older than the policy refresh interval.
   • Surfaces conflicts where two prior answers disagree.
   • Returns a draft for counsel review before the sales team sees it.
   Confidentiality gateAnswer library lives in your tenancy. The agent refuses and flags if no approved source exists.

   Human in the loopCounsel approves every answer that touches privilege, security posture, or AI usage claims.

Phase by phase.

The end-to-end path the agent runs. Every phase logs what it captured, what consent applies, and where a human gates the next step.

1. 01

   Intake

   Structured form classifies type, urgency, counterparty, jurisdiction. Privilege flag set if counsel directs.

2. 02

   Retrieval

   Agent pulls relevant policy paragraphs, prior matters, playbook clauses, and live regulator text.

3. 03

   Draft

   First-pass redline, memo, or response is generated with paragraph-level citations and confidence scores.

4. 04

   Refusal check

   If a section has no source above the threshold, the agent refuses that section rather than fabricate.

5. 05

   Counsel review

   Named lawyer accepts, edits, or rejects. Edits feed back into the evaluator-panel record.

6. 06

   External action

   Counsel sends the redline, memo, or filing. The agent never sends externally on its own.

7. 07

   Outcome capture

   Settlement, negotiated terms, or filed outcome is recorded against the original prediction.

8. 08

   Archival and retrain

   Closed matter is indexed for future retrieval. Quarterly retrain uses your matter history only.

The non-negotiables, spelled out.

Consent, security, accuracy, and residency. Explicit, auditable, and the same line items your CISO or GC already asks about.

What the numbers look like.

Figures we can point at. Every one carries a source you can verify.

What it replaces. What it augments.

How the Legal Agent supports counsel

It studies your policies and matters, drafts cited responses for review, and flags risk early.

The numbers, up front.

Custom enterprise licence. Per-seat pricing punishes the team that gives the agent to its business stakeholders for intake, which is exactly the use we want. Per-matter pricing punishes high-volume teams and creates the wrong incentive to skip the agent on small matters. The licence scales with corpus size, deployment mode, and the size of the senior-counsel evaluator panel we run for you. Implementation includes the playbook intake, the policy library indexing, and the first evaluator-panel calibration.