Skip to main content
Analysts at hedge funds and family offices spend 87% of their day on data gathering, copy-paste, and cross-checking. cf0 is the financial intelligence stack that collapses that work: ask a question, get a cited answer, ship a branded PDF. In one Lab thread an analyst can pull a on a name, swap the , run a sensitivity, draft a memo, and export the PDF in the firm’s brand template. The passage that anchors every figure is one click away. Dashed line from outputs back to source: every figure cf0 surfaces is cited back to the exact filing, page, and section.

What ships in production

Lab

The research surface. Plain English in, streaming charts, tables, and cited paragraphs out. The model picks the UI the answer needs.

Reports

Branded PDFs with numbered footnotes, a Sources Table, and a Key Assumptions table. Ready for IC.

Skills

Reusable workflows — DCF, LBO, comps, IC memos, earnings recaps — invoked with /skill-name from any thread.

What ships from Lab

Lab

Plain English in, streaming charts and tables and cited paragraphs out.

Reports

Branded PDFs with numbered footnotes, a Sources Table, and a Key Assumptions table.

SEC + global filings

Every major SEC form structured to section level, plus directory coverage across the global regulators. See coverage below.

Documents

PDFs, spreadsheets, decks, images — parsed and queryable in the same Lab thread.

Skills

Reusable workflows — DCF, LBO, comps, IC memos, earnings recaps — invoked with /skill-name.

Knowledge

Topics, notes, sources, and a force-directed graph that explain what cf0 has compiled and where every claim came from.

SEC and global filings coverage

cf0 ingests and structures every filing form an analyst is likely to pull, across ten regulator-backed markets — SEC EDGAR plus nine international regulators.

SEC forms (US)

FormWhat it carries
10-KAnnual report — financials, MD&A, risk factors
10-QQuarterly report — interim financials
8-KMaterial events — earnings, leadership, M&A
S-1IPO prospectus
DEF 14AProxy — exec comp, board, governance
13-FInstitutional holdings (quarterly)
N-PORTFund holdings
N-CSRFund certified shareholder reports
N-PXFund proxy voting records

Global markets

MarketRegulator
USSEC EDGAR
UKFCA / Companies House
CanadaSEDAR+
JapanEDINET
South KoreaDART
IndiaSEBI
AustraliaASX
Hong KongHKEX
ChinaCNINFO
BrazilCVM
US filings get full section-level extraction (Item 1A risk factors, Item 7 MD&A, etc.) so Lab can quote the exact passage a figure came from. International coverage is directory + filing-list today, with selected markets advancing to full section ingestion — Brazil is the most recent. See SEC filings for the full pipeline detail.

Trust

  • Every figure traces to the exact filing, page, and section. Click to verify.
  • Reports include numbered footnotes resolving to a Sources Table, plus a Key Assumptions table before every valuation.
  • Threads export as compliance-ready audit trails.
  • Org-scoped data, never used to train AI models.
See Security overview for the full posture.

Common analyst questions

In deterministic code templates, not AI inference. DCF, WACC, multiples, sensitivities, Monte Carlo — all run as tested code paths the model frames the question for and reads the result from. Same question, same inputs, same answer. See Guardrails → the model doesn’t do math.
It says so. From a real generated report: “EV/Sales requires enterprise value; since debt/cash and shares outstanding are not provided here, EV/Sales is not computed.” cf0 surfaces gaps rather than inventing numbers. See refuse to fabricate.
Yes. Every Lab thread exports as a Markdown or PDF audit trail capturing the full conversation, every tool call and its result, every assumption gate and how it was resolved, every citation with source link, timestamps, and the model version per turn. See Citations and audit trail.
No. Customer threads, documents, and reports never enter any training set. AI inference runs inside cf0’s cloud perimeter — your data does not leave the platform to reach a model. See Data governance.
cf0 orchestrates a frontier large language model with specialised sub-agents for retrieval and synthesis. Every turn is fully traced — system prompt, tools called, model inputs and outputs — and the exact model version per turn is captured in the audit trail so you can verify after the fact. See Observability.
Four ways. (1) Numbers are computed in deterministic code templates, not generated by the model. (2) Citations trace every figure to the exact filing page. (3) Audit trails export per thread. (4) Compounding context — every shared document, integration, and memory entry sharpens the next answer across sessions. ChatGPT does none of this for institutional research.
Next: Set up your account →
Last modified on June 5, 2026