Accelerator 10 · Legal
NDA Intelligence: clause risk, compare and redline
Reviews NDAs clause by clause against your legal playbook, compares two drafts side by side and proposes redlines, with every finding tied to the exact clause text.
The problem
- NDAs are high volume and low value, yet each one waits in the legal queue for days.
- Business teams sign without knowing which clauses depart from the company's positions.
- Lawyers repeat the same redlines on every counterparty's paper, and reviews vary between reviewers.
- Generic AI tools produce confident summaries with invented scores and no link back to the clause text, which a legal team cannot rely on.
What it does
- Splits the NDA into clauses in code and classifies each into a fixed taxonomy of 15 clause types (definition, exclusions, term, non-compete, remedies, governing law and more).
- Runs deterministic playbook checks (missing exclusions, non-compete, perpetual term, liquidated damages, foreign governing law, text aimed at AI reviewers) that the model cannot override.
- Asks the model for a structured verdict per clause, validated against a schema. Every verdict must quote the exact clause text, or it is repaired once and otherwise rejected.
- Computes the overall risk in code from a documented rubric, never from the model, and marks results provisional when clauses could not be assessed.
- Compares two NDAs by clause type, drafts replacement wording with a word-level redline, and exports the review as Markdown or JSON for the matter file.
Pick a synthetic NDA (or paste your own synthetic text), choose the jurisdiction playbook, and run Review, Compare or Redline. Every risk is quoted from the clause, and the overall rating is computed by a published rubric.
Live demo: NDA Intelligence
Synthetic data only · nothing you type is stored
Mode
Clause-by-clause risk against the playbook
Please use synthetic text only. Max 12,000 characters per NDA. Each clause is assessed separately, so results appear as they arrive.
Not legal advice — for review by your legal team. The demo playbook (0.1-draft (12 Oct 2026)) contains policy positions only, no case law. It must be reviewed and approved by qualified counsel before any real use.
Pick a sample (the aggressive NDA hides an instruction aimed at AI reviewers), choose a mode and run it. You will see the clauses as the code splits them, then each verdict as it arrives.
The demo first calls the Surge in-house AI server and falls back to a hosted open-source model if it is busy. The badge on the output shows which one answered. Limited to a few runs per visitor per day.
How it integrates
In your deployment the accelerator runs inside your environment, behind your single sign-on, and is connected to the systems below. You choose the hosting and the model.
Systems and APIs
- Document management (iManage, NetDocuments, SharePoint) to pull NDAs and file the review
- Contract lifecycle management (Icertis, Ironclad, DocuSign CLM, Agiloft) as an intake and approval step
- Legal mailbox intake (Exchange / Microsoft Graph) with triage to the business or a lawyer
- Microsoft Word add-in to apply redlines as tracked changes
- SSO (Entra ID / Okta) with roles for requesters, reviewers and approvers; audit log to your SIEM
Data we need for a pilot
- Your NDA playbook: preferred, fallback and escalation positions per clause type and jurisdiction, approved by your legal team
- 30–50 negotiated NDAs with final versions to calibrate the risk thresholds
- Routing rules: which risk levels can go back to the business and which need a lawyer
Where it runs
- On-prem or in your private cloud; NDA text never leaves your network
- Runs on an open-source model behind your own OpenAI-compatible gateway (vLLM / LiteLLM)
- Per-clause calls keep each request small, so 14B–32B models work without long JSON outputs
Open-source models we start with
- Qwen3-class or Llama-class 14B–32B instruct model with JSON / guided decoding
- Optional embeddings over your negotiated clauses for precedent search
Walkthrough
A short recorded walkthrough of the nda intelligence: clause risk, compare and redline: the input, the output and how a reviewer uses it.
Next accelerator: Resume screening & interview kit →Walkthrough video coming soon
We are recording a short walkthrough of this accelerator. Meanwhile, run the live demo above or book a working session.
Run the legal accelerator on your data
In a working session we map it to your process, systems and rules, and agree what a pilot must prove before it goes live.