Research written from what only your fund has.
Core reads your notes, the broker reports you receive and your book, then writes the finished piece.
- Morning brief: how the book moved overnight and why, before the open.
- The book in every answer: positions, exposures and NAV, from your own records.
- Your library: broker notes and channel checks, searchable the moment they land.
Every figure, back to its sentence.
Ask a question and Core answers with a full page, not a paragraph, and cites where each number came from.
- Guidance trackers: what a company guided, what it printed, and what your preview expected.
- Cited: click a figure to see the passage in the filing or note.
- Verified: figures Core computed are marked apart from figures it quoted.
Rumors checked against your own evidence.
Paste a headline or a post. Core weighs it against your channel checks, filings and data, and says what would change its mind.
- Verdicts: likely true, uncertain or likely false, claim by claim.
- Evidence: your analysts’ checks next to public filings and shipments data.
- What would change it: the signal to watch next.
Your data. Your model. Your machine.
Core runs on a machine in your office. You choose what leaves the firm, down to nothing at all.
- Your model: Qwen on your own NVIDIA hardware, or Claude, GPT or Gemini under your agreement.
- Accounted for: a model card, an answer log and an inspection pack in one download.
- Checkable: every outbound connection is listed on our security page.
- Model
- fine-tuned on your research
- Weights
- yours
- Hash
- sha256 9f3a…c21e
- Reviewed
- verified
- Core node
- running
- Location
- your office
- GPU
- NVIDIA, yours
- Outbound traffic
- none
Core vs Hebbia
Most rows follow what the SFC, MAS and FSC ask of AI in investment research: where it runs, what leaves the firm, a source for every figure, and a record of who reviewed it.
| Capability | Core | Hebbia |
|---|---|---|
| Runs on a machine in your office | ||
| Can run with nothing leaving the building | ||
| Open-weight model on your own GPUs (Qwen on NVIDIA DGX Spark) | ||
| Claude, GPT or Gemini under your own agreement | ||
| Reads your fund’s book and answers with it | ||
| Korean DART filings read in full | ||
| Reads Korean, Japanese and Chinese documents | ||
| Model card: which model ran, where, and its hash | ||
| Answer log: every answer, its sources, who reviewed it | ||
| Controls mapped to the Hong Kong SFC’s generative-AI principles | ||
| One-download inspection pack for auditors | ||
| A public list of every outbound connection | ||
| A paid pilot on your own documents and hardware |
Hebbia column from Hebbia’s public materials, checked 24 October 2026. “Not offered” is where those materials show it; a dash means they don’t say. Rows where both include it are left out. Sources below.
“After closely comparing five vendors, Treasurer delivered the most accurate figures and analysis.”
Treasurer works with
Frequently asked questions
Looking for Hebbia alternatives?
If your documents can’t go to a vendor’s cloud, Core runs on a machine in your office and can use an open model with nothing leaving the building. If you want a hosted platform for running questions across thousands of documents, Hebbia is strong there.
How is Core different from Hebbia?
Hebbia is a cloud platform for finance and law, on frontier models it provides. Core is installed in your office and writes fund research from your notes, broker reports and book, on the model you run or choose.
Can we use Core alongside Hebbia?
Yes. Core can handle the research that must stay in the building while Hebbia works across data rooms and large document sets in the cloud.
Does Core run on-premises?
Yes. Core runs on a Mac mini, an NVIDIA DGX Spark or your own GPU server in your office. On a workstation or your own GPU server, documents, model and chat all stay inside, with Qwen as the model.
Which fits the SFC, MAS and FSC rules on AI better, Core or Hebbia?
No regulator certifies software, so neither is "compliant" on its own; your compliance team decides. The rules ask for a person checking the output, confidential data kept inside the firm, and an accountable record. Core runs on a machine in your office, cites the source page of every figure, logs each Ask answer with its reviewer, and exports an inspection pack framed on the SFC’s principles. Ask Hebbia how it meets the same duties.
Why not use ChatGPT, Claude or Gemini directly?
You can, under your firm’s AI policy. MAS now expects a written rule on whether and how staff may use AI, to manage “shadow AI”. Core gives the desk one sanctioned tool: cited figures, an answer log and a machine in your office, and it can run Claude, GPT or Gemini under your own agreement.
