by
Core · Private AI for finance firms

Ask. It’s answered.

Answers from your notes, filings and licensed data, on a machine in your office.

A film of Core, on demo data. The morning Brief says the book was up 8 bps overnight, and a citation opens the analyst note behind it. Asked where our NVIDIA preview sits against consensus, Core searches S&P Global Visible Alpha and answers with a chart, then lists its six sources: licensed data, filings and the firm’s own notes. One source opens in the Library with the cited sentence highlighted. Then a rumor that Microsoft cut its GB300 order is weighed against the firm’s evidence: likely true, 84%.

Used by professionals at

Mirae Asset Global Investments, Timefolio Asset Management, Midas Asset Management, Samsung Active Asset Management, Daol Asset Management, Anda Asset Management, C-Square Asset Management, Mastern Investment Management, IMC Trading, Shinhan Securities, Hanwha Investment & Securities, Kyobo Securities, Hana Securities, NH Investment & Securities, Samsung Securities, Korea Investment & Securities, Daol Investment & Securities, DB Securities, UBS

Inside Core

One system.

  • Brief

    What moved your book overnight, written before the open.

  • Ask

    Ask in plain words. Every figure links to its sentence.

  • Library

    Notes, broker reports and filings, searchable as they land.

  • Rumor Check

    A post weighed against your notes, with how likely it is.

  • Guidance Tracker

    What a company guided, what it printed, and consensus.

  • Portfolio

    Positions, exposure and performance from your holdings.

  • Markets

    Indices, rates and currencies, for desks without a book.

  • Data Packs

    Optional: signed reference data on companies, people and deals.

What it does

It reads, fetches, writes and checks.

Knowledge graph

Read

Core reads every document you give it into a map of companies, funds and deals, each link kept with its sentence.

How the map works

A map of a firm’s documents in 3D: regions of companies, products, documents and claims, linked within and between them. One claim, an order cut, is traced from SK hynix and NVIDIA through the GB300 to the channel check that states it.

Virtual analyst

Fetch

Ask for a filing, a report or a paper. It finds it in a disposable browser, reads it, and leaves it for you to add.

How the virtual analyst works

Earnings Release

3Q26Oct 28PDFPDF
2Q26Jul 24PDFPDF
1Q26Apr 23PDFPDF
PDF3Q26_Earnings_Release.pdf✓

Brief and Ask

Write

The Brief before the open, and answers to any question, with every figure linked to its sentence.

Core for funds

Guidance tracker · NVIDIA · Revenue, $BCore · 13 Oct 2026

NVIDIA has printed above the top of its guided range for three straight quarters.

Each print landed above the ±2% range NVIDIA guided to NVIDIA Q4 FY2026 earnings release · Feb 2026NVIDIA Q1 FY2027 earnings release · May 2026NVIDIA Q2 FY2027 earnings release · Aug 2026. Our previews ran below the print both times NVDA Q1 FY2027 earnings preview · J. Park, internalNVDA Q2 FY2027 earnings preview · J. Park, internal; the Q3 guide steps up 18.7% on Q2’s NVIDIA Q2 FY2027 earnings release · Aug 2026, and our preview sits above it NVDA Q3 FY2027 earnings preview · J. Park, internal.

  • Printed above the range3 of 3ƒ
  • Our preview vs print(1.9%) avgƒ
  • Q3 FY27 guide$108.0BNVIDIA Q2 FY2027 earnings release · Aug 2026
  • Our Q3 preview$113.0BNVDA Q3 FY2027 earnings preview · J. Park, internal
Prints against the guided range · % of the guide midpoint
-4%0%+4%+8%Q4 FY26Q1 FY27Q2 FY27Q3 FY27
QuarterGuidedPrintedvs guideOur preview
Q4 FY26$65.0B ±2%NVIDIA Q3 FY2026 earnings release · Nov 2025$68.1BNVIDIA Q4 FY2026 earnings release · Feb 2026+4.8% ▲ aboveƒ—
Q1 FY27$78.0B ±2%NVIDIA Q4 FY2026 earnings release · Feb 2026$81.6BNVIDIA Q1 FY2027 earnings release · May 2026+4.6% ▲ aboveƒ$80.4BNVDA Q1 FY2027 earnings preview · J. Park, internal
Q2 FY27$91.0B ±2%NVIDIA Q1 FY2027 earnings release · May 2026$96.2BNVIDIA Q2 FY2027 earnings release · Aug 2026+5.7% ▲ aboveƒ$94.0BNVDA Q2 FY2027 earnings preview · J. Park, internal
Q3 FY27$108.0B ±2%NVIDIA Q2 FY2027 earnings release · Aug 2026—open$113.0BNVDA Q3 FY2027 earnings preview · J. Park, internal
  1. The beats are widening: +4.8%, +4.6%, then +5.7% against the midpoint NVIDIA Q4 FY2026 earnings release · Feb 2026NVIDIA Q1 FY2027 earnings release · May 2026NVIDIA Q2 FY2027 earnings release · Aug 2026.
  2. Our previews have been conservative, 1.5% and 2.3% under the print NVDA Q1 FY2027 earnings preview · J. Park, internalNVDA Q2 FY2027 earnings preview · J. Park, internal.
  3. Q3 is guided to $108.0B; our preview of $113.0B would be another beat above the range NVIDIA Q2 FY2027 earnings release · Aug 2026NVDA Q3 FY2027 earnings preview · J. Park, internal.
Data Center revenue · $B
Sources
  1. 1NVIDIA Q3 FY2026 earnings release · Nov 2025
  2. 2NVIDIA Q4 FY2026 earnings release · Feb 2026
  3. 3NVIDIA Q1 FY2027 earnings release · May 2026
  4. 4NVIDIA Q2 FY2027 earnings release · Aug 2026
  5. 5NVDA Q1 FY2027 earnings preview · J. Park, internal
  6. 6NVDA Q2 FY2027 earnings preview · J. Park, internal
  7. 7NVDA Q3 FY2027 earnings preview · J. Park, internal
✓Verified · 10 figures cited · 5 computed by Core · 7 sources

Rumor Check

Check

Paste a post. Core weighs each claim against your notes, filings and data, and says how likely it is.

Request a demo

A rumor check of a post saying Microsoft cut its GB300 rack order by about 20% and pulled in Maia 200: the order cut is likely true at 84%, the Maia 200 pull-in uncertain at 38%, and a risk to NVIDIA’s Data Center outlook likely false at 13%, with the channel checks and shipment data behind the first.

Who it’s for

Set up for your firm.

Run a sharper book with private AI

Kill criteria checked every day, theses written to your standard, and a Brief over the book before the open.

Core for hedge funds and asset managers

Core for funds: the portfolio overview.

Security and compliance

Private. Cited. Accountable.

  • Runs in your office

    On a DGX Spark or your own servers. Every connection it makes is listed, with how to turn it off.

    Security and compliance
  • Example model card: Qwen3.8, open weights under Apache 2.0, running on a DGX Spark in your office, used as published, with the hash of its weights recorded

    Your model, accounted for

    The model you choose, with a card your compliance team can read.

    Your model
  • Example answer log: each answer with its reviewer
    Answer logQwen3.8 · on-prem
    • 09:12NVDA guidance trackerReviewed · J. Park
    • 09:31Rumor check · GB300 orderReviewed · S. Lim
    • 10:04Peers on Q3 marginsAwaiting review

    Every answer on record

    Who asked, which model answered, where it ran and who reviewed it, exported in one download.

  • Built for the rules

    Mapped to what Hong Kong, Singapore and Korea ask of AI in investment research.

“After closely comparing five vendors, Treasurer delivered the most accurate figures and analysis.”
AI Lead, Data & Content · TheBell
Questions

What firms ask before they start.

What is Core?

Core is private AI for finance firms, built by Treasurer. Ask a question and it writes a complete report, tracker or check from your firm’s notes, documents, filings and licensed data. Every figure links to the sentence it came from. It is set up for hedge funds and asset managers, research houses, venture investors and reinsurance brokers.

Who builds Core?

Treasurer, the AI platform for finance, with offices in Seoul and Singapore. Treasurer builds the engine that collects, refines and reasons over financial data, and the products on it: AlphaLenz, ARIA and Findle. Core runs that engine in your office. Treasurer is a member of NVIDIA Inception, and Core runs on NVIDIA hardware, from a DGX Spark to your own GPU servers.

Who is Core for?

Finance firms whose edge is in their own documents, which can’t go to a public AI: funds and asset managers, research houses and publishers, venture and private equity investors, and reinsurance brokers. It reads Korean DART filings in full as well as US filings, in Korean, English, Chinese and Japanese, so it suits firms that work in Asia.

Does our data leave the office?

It depends on how you run Core. On a DGX Spark or your own GPU server with an open model, your documents, questions and answers stay on your network. In our private cloud, your documents sit on a server we run for your firm alone, and text goes to the model provider. The security page lists every connection Core makes, with how to turn it off.

Which AI models does Core use?

Qwen on an NVIDIA DGX Spark or your own NVIDIA GPUs; or Claude, GPT or Gemini under your own agreement with Anthropic, OpenAI or Google. The model card records which model ran, where it ran, and its hash.

Can we use Core from Claude or ChatGPT?

Yes. Core is an MCP server: each person connects their own Claude, ChatGPT, Cursor, Codex or Gemini CLI with a personal token, and it searches your library as them, with their access. See treasurer.net/core/api.

What hardware do we need?

One of three: an NVIDIA DGX Spark (GB10, 128 GB) on a desk, which we can supply; an NVIDIA GPU server your IT team already runs, sized with us; or none at all, in a private cloud server we run for your firm.

How does Core help with the SFC, MAS and FSC rules on AI?

All three ask for the same things: a person checks the output, confidential data stays inside the firm, and someone is accountable for a record. In Core every figure cites its source page and a figure with no source is flagged. Each Ask answer is logged with the model, where it ran, its sources and who reviewed it. The model card, controls, use-case inventory and answer log download as one inspection pack, framed on the SFC’s principles.

Is Core certified by the SFC, MAS or FSC?

No regulator certifies software, and none has certified Core. Core is built for the duties they set out. Whether your use needs a notice to the SFC, or a sandbox exemption in Korea, is a call for your compliance team. We walk them through the machine, every outbound connection and the inspection pack before anything is installed.

Who reviews an answer, and is that recorded?

Your analyst or portfolio manager. Each figure links to its source page, so checking a number takes one click. Each Ask answer is logged with the reviewer’s name, the model, where it ran and its sources.

Where are the records kept?

On your own machine, in your office. The answer log is not stored in a provider’s cloud abroad, which is what the SFC’s rule on records kept with cloud providers is about. In our private cloud, records sit on the server we run for you, and we keep a full copy in your office if you choose.

How do we get started?

A 30-minute demo, then we map your sources and your machine and send a proposal. Set-up installs Core on your hardware, or on a private cloud server, and connects your documents. Your IT and compliance teams get a walk-through of the machine, every outbound connection and the inspection pack before anything is installed.

How much does Core cost?

Every Core is scoped to your firm: a set-up fee and a monthly partnership fee, in a written proposal after the demo. It isn’t priced per seat, and questions on your own machine aren’t metered. Hardware, cloud-model tokens and add-ons are separate, and there is no public price list. See treasurer.net/pricing, write to contact@treasurer.net, or request a demo on this page.

NVIDIA Inception ProgramNVIDIA Inception member
treasurer.net ↗

© 2026 Treasurer Global Pte. Ltd. All rights reserved.

Member of the NVIDIA Inception Program