Skip to main content

Extract data from NDAs

An NDA is the contract that lets two parties exchange confidential information before they have a longer-term relationship. Sales teams sign one before a vendor demo. Investors sign one before a data room. M&A targets sign one before due diligence. Most enterprise legal departments handle thousands of NDAs a year, and the operational pain is that they pile up in inboxes, SharePoint folders, and contract management systems with inconsistent metadata, so when a question arises (when does the NDA with Acme expire? was it mutual? does it cover residuals?), someone has to open the actual PDF to find out. Procurement and BD teams need the same metadata for a different reason: tracking which counterparties are under an NDA, when each expires, and whether the scope covers a specific evaluation that has come up. The hard parts are the variations within a deceptively simple structure. NDAs come as mutual (both parties exchange info) or unilateral (only one party discloses). Term lengths vary from 1 year to perpetual. Definition of Confidential Information ranges from "all non-public information" to a multi-paragraph carve-out list. Permitted uses are often restricted to a named evaluation (e.g., "evaluating a potential investment in the Company"). Standard exceptions cover publicly known information, independently developed information, and information received from a third party. Residual knowledge clauses (the receiving party's employees can use general skills and concepts they learn during the engagement) are highly negotiated. Return-or-destroy obligations specify what happens to the information after the engagement ends. Governing law and dispute resolution venue are negotiated heavily by sophisticated counterparties. Talonic extracts the structured metadata while preserving the clause text. Parties are tagged with their role (disclosing, receiving, both for mutual NDAs). Term length, governing law, and dispute resolution venue surface as discrete fields. Confidentiality scope, residual-knowledge language, return-or-destroy obligations, and named evaluation purpose are captured as clause excerpts so legal can scan the actual text rather than rely on a parsed approximation.

What gets extracted from NDAs

Contract TitleMutual Non-Disclosure Agreement
PartiesAcme Software, Inc.; Globex Logistics LLCMutual or unilateral
Effective Date2026-04-05
Term Description3 years from the Effective Date
Confidentiality ScopeAll non-public information disclosed by either party
Permitted UseEvaluation of a potential commercial partnership
Governing LawState of Delaware
Dispute Resolution VenueState or federal courts in New Castle County, DE

How extraction works for NDAs

NDAs differ heavily by who drafted them: a vendor's standard mutual NDA, an investor's unilateral NDA, an enterprise's procurement NDA with embedded data-protection terms. Talonic classifies the NDA (mutual vs unilateral, standalone vs included as an article of a larger agreement) and runs it through the contract schema in the Field Registry, which captures parties, term, confidentiality, governing law, and termination. Defined Terms (Confidential Information, Receiving Party, Disclosing Party, Permitted Use) are resolved in context so the extracted scope reflects what the contract actually means. Return-or-destroy obligations and residual-knowledge clauses are captured as clause excerpts because the negotiation around them is too varied to standardize. Per-cell confidence with pixel-region provenance follows DIN SPEC 91491 conformity, so counsel can audit any captured clause against the original NDA before relying on the metadata for renewal tracking or scope verification.

Sample extraction

A standard mutual NDA between two software companies

{
  "contract_title": "Mutual Non-Disclosure Agreement",
  "contract_type": "nda",
  "parties": [
    {
      "name": "Acme Software, Inc.",
      "role": "mutual"
    },
    {
      "name": "Globex Logistics LLC",
      "role": "mutual"
    }
  ],
  "effective_date": "2026-04-05",
  "execution_date": "2026-04-05",
  "term_description": "3 years from the Effective Date; confidentiality obligations survive termination for an additional 2 years",
  "confidentiality": "All non-public information disclosed by either party in any form, including business plans, customer lists, technical specifications, and financial information",
  "permitted_use": "Evaluation of a potential commercial partnership between the parties",
  "governing_law": "State of Delaware, USA",
  "dispute_resolution_venue": "State or federal courts located in New Castle County, Delaware",
  "termination": "Either party may terminate on 30 days written notice; confidentiality obligations survive"
}

Frequently asked

Does it distinguish mutual from unilateral NDAs?

Yes. Each party is tagged with a role (mutual, disclosing-only, receiving-only). A mutual NDA returns both parties as "mutual"; a unilateral NDA returns one as disclosing and one as receiving. The party_count and role tags drive downstream filtering when a legal ops team needs to find all NDAs where their company is the disclosing party.

How are term length and survival periods captured?

Term description is preserved verbatim because the language varies ("3 years from the Effective Date", "until terminated by either party on 30 days notice", "perpetually with respect to trade secrets"). When the language is unambiguous, a structured term_months field is also populated. Survival language (e.g., "confidentiality survives termination for 5 years") is captured separately.

What about defined terms like Confidential Information?

The definition is captured as a clause excerpt rather than a parsed boolean of what is covered. NDA scope is too variable to summarize into structured fields without losing meaning; legal review of the verbatim language is the correct downstream step.

Can it handle NDAs embedded inside a larger agreement?

Yes. When confidentiality is an article of a Master Service Agreement or an asset purchase agreement rather than a standalone NDA, the parent contract is classified as the larger agreement type and the confidentiality article is captured as one of its clauses. The structured fields surface the parent contract metadata.

Are residual-knowledge and return-or-destroy clauses flagged?

Yes. Both are highly negotiated and matter to downstream consumers (security, compliance, legal). The clause text is captured verbatim with a clause-type tag so a search across the contract repository can find every NDA with a residual-knowledge carve-out or a return-or-destroy obligation.

Author note

Reviewed by Talonic engineering, contract schema review · last reviewed 2026-05-10