Skip to main content

Guide

Is Aventro just a ChatGPT wrapper?

No, though it does use OpenAI’s models to read documents and draft text, and Microsoft’s to transcribe call recordings, all through Vercel AI Gateway with zero data retention on every call. What counts as verified, which record matches the company, what the returns are, where documents disagree and who can see what are decided in code, not by the model.

Book a demo

By , co-founder & CTO · Updated . Sources are listed at the end.

What is code, not a model

The parts of Aventro a prompt cannot change, each described on the page linked.

Checks against 35 published sources

Company registers in Israel, the UK, the US, Norway and Finland, GLEIF, SEC EDGAR, patent offices and more, each queried directly. An official register outranks a commercial aggregator, which outranks the open web.1,2

Every source and its weight

Entity matching the compiler enforces

A result cannot be marked verified unless it carries the match that justifies it, such as a registration number, an LEI or an SEC filing, and the type system refuses it otherwise. A matching name is shown as a lead.3,2

How verification works

Contradiction detectors

Plain code compares each figure across the deck, the documents, the accounts and what was said on calls, and raises a question when they disagree. No model decides that two numbers differ.4

Contradictions, step by step

The returns engine

Your ownership, diluted round by round to exit, across five scenarios from failure to a home run, computed with no model call and with every base rate cited beside the number it produced.5

Exit and returns

The scoring rubric

The Aventro score is a weighted roll-up of modules and sub-factors. Some sub-factors are computed from the company’s figures, a model assesses the qualitative ones against a rubric, and missing data is left out rather than scored as zero.6

How screening scores

Versioned Fund DNA

Your mandate as criteria, saved as versions under the name of the person who saved them. Every Fund Fit names the version it used, so a change never rewrites an old score.7

Fund DNA

Security and access

Evidence encrypted with AES-256-GCM under per-deal keys, reach enforced in the database per deal and per person, and sensitive actions written to a security audit log.8,9

Security and your data

One record per company

However a company reaches you, it lands as one opportunity, and the deck, the documents, the checks, the company’s answers and your decision stay on it.10,4

Opportunities

What the model does

Four jobs, each with a limit around it.

  1. Reads documents

    Decks, data rooms and call transcripts are read into claims and figures, which the code then checks.

    In Aventro: PDFs, PowerPoint decks, spreadsheets, Word files, CSV, images and call transcripts, up to 50MB each, uploaded by the fund or the company. Each document is read as it arrives, filed, and turned into checks.4

  2. Drafts

    Requests to the company, memo sections, comparisons and fields.

    In Aventro: Nothing Aventro AI drafts is saved until a person confirms it, and it is recorded as that person’s act.11

  3. Assesses the qualitative factors

    Team, market and product judgments that a formula cannot make.

    In Aventro: Each sub-factor ties back to the evidence behind it, and the report labels every section for how it was made, computed or AI-assisted.6,12

  4. Answers questions

    Across your pipeline, one company or the market.

    In Aventro: Aventro AI reads exactly what you can read and nothing more, and its answers name the records they came from.11

Questions

Which models does Aventro use?

OpenAI’s models to read and draft, Microsoft’s to transcribe call recordings, and TypeSafe’s for short typed readings, all through Vercel AI Gateway. Every call runs with zero data retention, so the provider keeps nothing once it has answered, and the gateway chooses the region for each request rather than pinning one. Every subprocessor is listed on the subprocessors page.

Are our documents used to train models?

No. Aventro does not authorize its AI providers to train on your content, and its terms say so.

Could we build this ourselves with ChatGPT?

You could get a model to read a deck. The work is everything around it: an integration with each register, matching rules for each source, the contradiction detectors, the returns engine, versioned mandates, permissions enforced in the database, and the record your team keeps. That is what Aventro is.

Does Aventro make the investment decision?

No. Aventro screens, checks and drafts. A person on your team records the decision, with their own reasons, and Aventro does not give investment advice.

Bring one live deal

Free, on one of your own deals. See it on a company you already know, before any quote.

  • A live deal screened against your own Fund DNA
  • Which claims we flagged that your team had not reached yet
  • How claims get checked against official registers and public filings
  • Your own IC memo template, filled from the deal
  • What pricing looks like for your seats and deal volume
Book a free demo

Sources

Every Aventro fact on this page links to the page of this site that states it.

  1. 1Aventro, Verification sources. aventro.ai/verification
  2. 2Aventro, Docs: Verification. aventro.ai/docs/verification
  3. 3Aventro, Blog: How an AI due diligence tool should verify a company. aventro.ai/blog/how-ai-due-diligence-should-verify-a-company
  4. 4Aventro, Docs: Diligence. aventro.ai/docs/diligence
  5. 5Aventro, Docs: Exit and returns. aventro.ai/docs/exit-returns
  6. 6Aventro, Docs: Screening. aventro.ai/docs/screening
  7. 7Aventro, Docs: Fund DNA. aventro.ai/docs/fund-dna
  8. 8Aventro, Docs: Security and your data. aventro.ai/docs/security-and-data
  9. 9Aventro, Docs: People and permissions. aventro.ai/docs/people-and-permissions
  10. 10Aventro, Docs: Opportunities. aventro.ai/docs/opportunities
  11. 11Aventro, Docs: Aventro AI. aventro.ai/docs/aventro-ai
  12. 12Aventro, Docs: Reports and sharing. aventro.ai/docs/reports

Related