Five focused agents. One attorney gate.

The first line of contract review and negotiation for in-house counsel. Five bounded agents do the first-pass work. A licensed attorney signs every output before it returns to your legal team.

01 — THE PIPELINE

One contract, six stages.

  1. Received
    Your request lands in the queue with all uploads and instructions captured.

  2. AI Review
    Our AI engine redlines against your playbook and flags risk.

  3. Attorney QA
    A CyberSquire attorney reviews the AI output, refines redlines, and writes the issues memo.

  4. Redlines Ready
    Markup-ready Word doc, clean version, and plain-English issues memo posted to your matter.

  5. In Negotiation
    Optional. We support negotiation rounds, counter-redlines, and live calls when you escalate.

  6. Closed
    Final executed version archived with full version history.

02 — THE AGENTS

Five agents. One job each. One way out.

INTAKE

First touch.

Classifies every contract, runs OCR and completeness checks, predicts risk, and routes the matter to the right playbook. Asks clarifying questions inside the matter when an upload is incomplete.

Stops at: Hand-off to redline.

REDLINE

First-pass redline.

Marks up the contract against your playbook — preferred clauses, fallback ladder, walk-away terms — and writes clause-level explanations with green/yellow/red risk tiers.

Stops at: Hand-off to QA. Never delivers to you.

ISSUES MEMO

Plain-English summary.

Translates the redline into a memo for business stakeholders — what changed and why, what's still open, and the items requiring attorney attention.

Stops at: Hand-off to QA.

QA

Adversarial cross-check.

A different model reviews the redline and memo for hallucinations, missed clauses, citation errors, and playbook drift. On fail, returns the matter to the redline step with specific corrections.

Stops at: Mandatory attorney review.

NEGOTIATION

Strategy on escalation.

When you escalate, prepares the attorney with counterparty patterns, a concession ladder, counter-redlines, and a call agenda. The attorney runs the negotiation; the agent is a strategist.

Stops at: Attorney runs the deal.

Agents have bounded tool access, cannot deliver work to clients, and cannot override your playbook.

03 — THE MODEL STRATEGY

Multi-model on purpose.

We never tie our pipeline to a single model. Each agent runs on the model best suited to its job, and the QA agent always runs on a different model family than the redline agent — so independent systems catch each other's errors.

  • Intake — small, fast model for classification and routing.
  • Redline — frontier reasoning model for long-context legal drafting.
  • Issues memo — frontier model tuned for synthesis and tone.
  • QA — different model family from Redline, for adversarial cross-check.
  • Negotiation — frontier reasoning model for strategy and counterfactuals.

Why cross-model QA matters

When the same model that drafted a redline also reviews it, it tends to rationalize its own mistakes. Routing QA to an independent model family is the single highest-leverage hallucination control we deploy.

04 — GOVERNANCE

Three points where a human always intervenes.

Pre-AI

Playbook setup. A CyberSquire attorney builds the playbook with you. Every later AI action sits inside the judgment captured at this stage.

Post-QA

Mandatory attorney review. Every matter, every output — attorney-reviewed and signed before it returns to in-house counsel. No exceptions. This aligns with ABA Formal Opinion 512.

Escalation

Live attorney work. When you escalate, or when QA flags a high-risk item, an attorney runs the matter directly.

05 — PRIVACY

Built by a privacy lawyer.

Zero data retention.

Your contracts are never retained, logged, or used to train upstream models. We sign zero-data-retention agreements with every model provider before we use them.

Tenant isolation.

Each client's playbook, embeddings, and matter history are isolated. No cross-client retrieval. Full stop.

Audit-grade logging.

Every upload, message, agent step, attorney decision, and download is logged in the Work Center with timestamps and actor identity.

Encryption everywhere.

TLS 1.3 in transit, AES-256 at rest, role-based access in the Work Center.

06 — THE BOUNDS

Bounded by design.

What agents cannot do

  • Agents do not deliver work to clients. Only an attorney does.
  • Agents cannot override your playbook. Every redline traces to a playbook rule or a documented exception.
  • Agents cannot communicate with counterparties.
  • Agents cannot accept new work outside the Work Center.

Ready to see the agents work?

Book a 20-minute consult. We'll walk through a real matter end-to-end.