Deal data, walled by fund and by deal.
Private equity firms hold confidential information from targets, portfolio companies, co-investors, and limited partners, often under NDAs that restrict who may see it. Data.co Control applies those walls to AI so a deal team's assistant sees its deal and nothing else.
NDAs and nonpublic information do not forgive a shared index.
A firm evaluating two competing targets, or holding public-company information through a portfolio board seat, cannot let a single AI index blend those sources. Confidentiality agreements often name the individuals permitted to see materials. Advisers also need to show controls around material nonpublic information.
Data.co tags every document with its fund, deal, and source agreement at ingest. Policies tie access to team membership, and the retrieval layer filters results before the model sees them. When a deal closes or dies, its wall can be closed for good, with the audit trail of who accessed what retained.
Walls that follow the deal.
- 1
Open
A new deal gets a tag, a team group, and a data room connector. Only that group, and agents acting for it, can read the deal's documents.
- 2
Diligence
An assistant summarizes CIMs, quality-of-earnings reports, and contracts with citations, scoped to the deal.
- 3
Committee
IC memo drafts pull from the deal and the fund's own policies. Cross-deal comparisons require an explicit approval.
- 4
Close or exit
Closed deals move to portfolio scope. Dead deals are sealed, and materials can be purged according to NDA return-or-destroy terms.
Where AI helps without crossing walls.
Diligence Q&A
Ask questions across a single data room and get answers with document and page citations.
Portfolio reporting
Normalize monthly KPI packages from portfolio companies, scoped to the operating partners assigned to each.
LP request responses
Investor relations drafts responses from fund documents and LP records, which deal teams cannot read.
Comparable analysis
Analysts compare against prior deals only where policy allows, and every cross-deal read is logged.
Clean-team arrangements as policy.
Where a clean-team agreement limits competitively sensitive information to named individuals, those names become a group and the restriction becomes a deny rule for everyone else, including agents.
Team membership decides retrieval.
A single rule ties deal documents to the deal's team group. Because the default is deny, a new deal is walled the moment its documents are tagged, before anyone writes a rule for it.
id: pol_pe_deal_001
description: Deal documents are readable only by that deal's team
subject: { type: [user, agent] }
resource: { collection: "dealroom.*" }
action: [read, search]
when:
subject.groups: { contains: { ref: resource.attr.deal_team } }
effect: allow
explain: "Deal materials are limited to the staffed deal team."Illustrative example · final API surface documented in /docs
What firms connect, and what governs it.
| Source | Typical obligation | Default scope |
|---|---|---|
| Virtual data rooms | NDA with named recipients, return-or-destroy | Deal team |
| Investment committee materials | Internal confidentiality, fund mandate | Fund team and committee |
| Portfolio company reporting | Shareholder and board agreements | Assigned operating partners |
| Board materials from public holdings | Possible material nonpublic information | Named board designees |
| LP records and side letters | Investor confidentiality, data privacy | Investor relations |
What your CCO and general counsel will ask.
- Can a document's embeddings and derived summaries be purged with the document?
- Is a dead deal sealed for every identity, including agents and administrators?
- Are clean-team names enforced as deny rules rather than training guidance?
- Can we show which identities accessed a data room during a given week?
- Do cross-deal comparisons leave an approval record?
Questions buyers ask
Can the same model serve every deal team?
Yes. Isolation comes from retrieval and tool policies, not separate models. The model only ever receives context that the requesting identity is allowed to see.
How do we handle NDA return-or-destroy obligations?
Deal-tagged documents, their embeddings, and derived summaries can be identified and purged together. The audit log retains the record of access without the underlying content.
Do portfolio companies need to install anything?
No. Portfolio data arrives through the reporting channels you already use and is tagged at ingest. Firms that also offer AI to portfolio companies typically deploy separate instances.
Does deal data leave our environment?
Not through Data.co. Inference, retrieval, and audit run inside your environment, and app.data.co handles only accounts, licensing, and billing.
What deployment model fits a mid-sized firm?
Most firms start with a private cloud tenancy or a small on-prem GPU server sized for diligence summarization and Q&A. The infrastructure assessment estimates load from document volume and the number of concurrent deal teams.
How are board seats at public companies handled?
Materials received through a public-company board seat can carry an MNPI label and a named-designee group. Policies deny everyone else, and the firm's compliance team decides when the label can be lifted.
See it against your own use case.
A 30-minute technical walkthrough, policy simulation on your scenarios, and a deployment recommendation for on-prem, private cloud, or hybrid.