Private beta · by invitation
The command layer that reads the raw room and returns a cited, committee-ready memorandum — versioned, audited, and yours alone.
Family offices were the first Houses we onboarded. The same closed system now serves venture, private equity, credit, and corporate development teams who hold themselves to the same evidence standard.
The same system is in use across venture, private equity, credit, corporate development, and strategic finance desks.
Structured artefacts your investment committee, counsel, and auditors can sign. Every claim points back to a source span, by hash.
Structured narratives with executive summary, risks, and IC questions — derived from the same source spans as the full memorandum.
25–40 page memoranda with LPA redlines, fee reconciliation, and track-record extracts.
Condensed memoranda with an explicit section on where you disagree with the lead sponsor.
Risk registers keyed to borrower materials — covenant exposure, collateral, and counterparty concerns.
Committee-ready materials for board reviews, financing decisions, and strategic updates.
Secondary pricing memoranda, concentration analyses, and NAV reconciliations.
Typed exception lists from ODD and diligence workflows — severity, rationale, follow-up questions.
Hash-chained records of every read, retrieval, model call, and memorandum lock on the engagement.
A chat tool hands finance teams an empty text box. Diligence OS starts with the workflow, the evidence standard, the review path, and the artifact that must survive partner scrutiny.
| Dimension | Generic chat | Diligence OS |
|---|---|---|
| Starting point | Chat-first | Workflow-first |
| Process | User-defined prompts | Versioned process |
| Provenance | Weak provenance | Source-span citations |
| Review | Hard to review | Partner review states |
| Outputs | Inconsistent outputs | Repeatable artifacts |
| Memory | No durable workflow memory | Append-only audit trail |
| Committee use | Difficult to defend | Committee-ready outputs |
Starting point
Generic chatChat-first
Diligence OSWorkflow-first
Process
Generic chatUser-defined prompts
Diligence OSVersioned process
Provenance
Generic chatWeak provenance
Diligence OSSource-span citations
Review
Generic chatHard to review
Diligence OSPartner review states
Outputs
Generic chatInconsistent outputs
Diligence OSRepeatable artifacts
Memory
Generic chatNo durable workflow memory
Diligence OSAppend-only audit trail
Committee use
Generic chatDifficult to defend
Diligence OSCommittee-ready outputs
§03 · The grounding
Retrieval grounds every claim in your own documents. When the evidence is not there, the system declines the claim rather than guessing. A memorandum you can defend is one that refuses to assert what it cannot support.
Every conclusion is bound to the sources that corroborate it. Brighter edges carry more support; a claim the room cannot substantiate is held back and marked — not softened, not invented.
Insufficient evidence — not asserted
Diligence OS does not replace judgment. It compresses the evidence work, standardizes the process, and produces decision-grade artifacts senior professionals can review, challenge, approve, and sign.
Admit sources
Documents enter through upload or connector import. Each file is scanned, hashed, and scoped to the engagement chamber.
Structure the review
The workflow DAG defines what runs, in what order, and what each stage depends on — not what the analyst remembers to ask.
Extract evidence
Retrieval reads source chunks by BM25 and vector fusion. Every extract cites the span it came from.
Register risks
Risks are typed, ordered by severity, and keyed to rationale citations — not free-text bullet points.
Draft artifact
The memorandum composes from upstream node outputs. A draft that fails schema validation is refused — no partial document.
Review and challenge
Partners review draft states, challenge claims, and rerun individual review steps without restarting the engagement.
Lock and sign
The locked version is immutable. Exports, share links, and audit events reference that version — not a moving target.
Each engagement runs a versioned workflow drawn from years of private- and public-market practice. The intelligence is the workflow and the evaluation harness.
IC memo
Allocators, family offices, fund-of-funds
Fee schedules, track records, and LPAs are read together — contradictions surface before the committee meets.
Manager IDD memorandum, 25–40 pages, structured to your House's reading template.
Investment committee pack, derived from the same source spans as the memorandum so the two cannot drift.
Each engagement runs inside a dedicated chamber. Inference moves through enterprise contracts that retain nothing. Every material action is logged against the engagement.
Each engagement runs inside a dedicated chamber. Documents, prompts, and outputs do not leave it.
Row-level security scopes every query to the House. Cross-tenant retrieval is forbidden by construction.
Sources are admitted into an envelope encrypted with a key the House controls. Firm-tier engagements may run open-weight models inside the House's own network.
Documents are never used to train any model. Ever. By anyone. The database refuses to mark a chamber otherwise.
Every claim in a memorandum points back to a source chunk by hash. If the underlying source changes, the citation flags red.
Every read, retrieval, model call, and lock is recorded against the engagement and hash-chained into an append-only ledger.
Default inference routes through enterprise contracts that retain nothing. Optional sovereign deployment for customer-operated chambers.
Built to the SOC 2 control set from day one; Type I observation has been underway since March 2026. The architecture — single-tenant chambers, House-held keys, zero-retention inference, append-only audit — does not depend on the certificate to be true. The auditor's name appears in the subprocessor list when contracted.
The standard
Before a workflow reaches a paying House, its latest run on a sealed dataroom must clear each of these. They are the bar we use to decide whether a workflow is ready, not a warranty on any one engagement.
The live numbers, workflow by workflow — including those still under tuning during the beta — are published on the principles page.
A six-week diligence, conducted to one standard, in a long weekend — without compressing the evidence.
During the private beta we onboard fewer than a dozen Houses, hand-in-hand with the founder. There is no public sign-up. Each introduction is reviewed before product access.
Workflow mapping
We map your diligence practice to the eight workflows — which run today, which wait for the next cycle.
Source format review
PDF, DOCX, XLSX, PPTX, and EML are accepted end-to-end. We review your typical data-room composition before the pilot.
Security posture review
Chamber kind, key posture, data residency, and MNPI routing are agreed before any document enters the system.
Pilot engagement
One live engagement on a real opportunity — staffed hand-in-hand with the founder, not a self-serve sandbox.
Team onboarding
Principals, analysts, and compliance reviewers learn the review console, citation reader, and lock workflow.
Output calibration
Memorandum templates, risk severity scales, and IC question formats are tuned to your House standard.
Partner review loop
The first locked memorandum goes through your partner sign-off process — we adjust until the artifact survives scrutiny.
Expansion across workflows
Once the pilot holds, additional workflows open in priority order — manager DD, direct IDD, ODD, and beyond.
Bring controlled AI workflows into your next capital decision.
No public signup. Introductions are reviewed before product access.