An analyst covering semiconductors the morning after NVIDIA files its 10-Q doesn’t need to be told revenue went up. The headline number hit their terminal before they finished the coffee. What they need is the why(s) and the what(s) buried in the footnotes: Did China revenue actually recover, or did the company just stop shipping there? Is the gross margin jump real operating leverage, or a one-time charge that didn't repeat? How concentrated is the customer base now, and is that getting better or worse?
None of that lives in a clean numeric feed. The answer is in the filing, but the filing is 90 pages long.
It lives in the filing text: in Management's Discussion and Analysis, in the segment footnote, in a single sentence in "Recent Developments" that quietly reframes the whole quarter.
So their real workflow begins: open EDGAR, find the right filing, Ctrl-F for "China," scroll past the boilerplate, cross-reference the segment note against the geographic table, copy the relevant lines into a doc, and manually note where each one came from so the individual can defend the number in tomorrow's morning meeting. Multiply that by every name they cover, every quarter, and it becomes painful.
The AI tools that were supposed to fix this often make it worse. A general-purpose chatbot will happily summarize "NVIDIA's latest quarter". However, it might be summarizing last year's filing, or hallucinating a number, and it almost never tells you which sentence on which page it's standing on. For anyone whose work has to survive compliance review or a skeptical PM, an unsourced answer is worse than no answer. You can't put "the AI said so" in an investment memo.
What AlphaCreek is and why it's different
AlphaCreek turns the entire universe of SEC filings into a structured, AI-queryable knowledge base without flattening the language into a spreadsheet.
Coverage spans annual reports (10-K), quarterlies (10-Q), foreign issuer filings (20-F and 6-K), and material-event disclosures (8-K). New filings are ingested as they hit EDGAR, so the data an analyst pulls reflects the company's most recent disclosure.
The key difference is structure plus provenance. Rather than handing back a wall of text, AlphaCreek parses each filing into a navigable table of contents and breaks it into addressable nodes right down to an individual paragraph or table inside a footnote. Every node carries a stable identifier and a citation URL that links straight back to that exact passage in a clean reader. That means an AI agent can fetch the precise paragraph, quote it, and attach a link a human can click to verify in one second. The result is the thing that's almost impossible to get elsewhere: an answer with a paper trail.
Here's a concrete example. Open NVIDIA's most recent 10-Q, and a numeric data feed will show that China revenue fell. What it won't tell you is the reason sitting in the prose: zero Data Center Hopper products shipped to China during the quarter, versus $4.6B a year earlier (source), while the geographic table shows China (including Hong Kong) revenue at $4.55B, down from $9.66B (source). And buried in "Recent Developments" is the detail that actually frames the next few quarters: starting February 2026, the U.S. government granted licenses to ship small amounts of H200 products to specific China customers, but no revenue has been recognized under that program yet, and any unit must clear a U.S. inspection and carry a 25% tariff on re-import (source). That's three load-bearing facts a number alone would hide. Now with the ability to trace it down directly back to the source, to the exact paragraph and page location.
AlphaCreek on OpenBB
The AlphaCreek App is now available on the OpenBB App Marketplace inside OpenBB Workspace.
The app includes two widgets:
The first is a Get Started widget that shows users how to connect AlphaCreek and ask questions with OpenBB Copilot or any other agent you brought in. It includes simple example prompts so analysts can immediately start exploring filing content.
The second widget shows AlphaCreek’s latest ingested filings, giving users a live view of recently updated SEC documents with direct links to the underlying source material. That makes it easy to see that the data is current and continuously refreshed.
Sources that are easy to verify without leaving your dashboard
OpenBB Copilot is working directly from the source, not a scraped summary. Every cited answer includes a link straight to that exact paragraph in the filing. Users can also open an iframe widget and embed the source document directly into their dashboard. This especially matters for analysts who need to defend every number, claim, and conclusion. Instead of copying text from a PDF and manually tracking where it came from, users can ask Copilot a question, get a sourced answer, and click straight through to the relevant passage in the filing.
Data that you can combine with other sources and your own research
AlphaCreek slots into OpenBB as another data source you can wire into your existing workflows. Pull a customer-concentration disclosure from AlphaCreek, line it up against your internal revenue-by-client data, then compare both to factor exposures or broker estimates, all in the same Workspace. SEC text stops being an isolated research step and becomes composable input, something you combine with your models, internal notes and research.
And because OpenBB allows users to bring their own agent into the Workspace, AlphaCreek will work perfectly not only with the native Copilot, but with any agent your firm has built and connected to the Workspace.
The workflow as a semiconductor analyst
An analyst covering semiconductors opens OpenBB Workspace, makes sure AlphaCreek is connected, and asks Copilot a single question: “Pull NVIDIA’s most recent 10‑Q and walk me through the major catalysts.” Within seconds, Copilot comes back with a sourced list of growth drivers and risks, each one tied to a paragraph‑level link in the filing.
The answer starts with the core engine: revenue of 81.6 billion dollars, up 85% year over year and 20% sequentially, and Data Center at 75.2 billion (+92% YoY), driven by Blackwell systems and networking attach. It then walks the analyst through the product cycle (Blackwell ramp today, Rubin as the next leg), the changing customer mix (roughly half of Data Center now coming from non‑hyperscalers), and the uncomfortable flip side: three direct customers still account for 21%, 17%, and 16% of total revenue, so a handful of build‑out cycles can swing the print.
From there, Copilot surfaces the constraints and balance‑sheet angles: AI infrastructure bottlenecks around data centers, energy, and capital; 119 billion dollars of manufacturing and capacity commitments, with 95 billion due by the end of fiscal 2027; and gross margin at 74.9% vs 60.5% a year ago, largely because last year’s quarter ate a 4.5 billion dollar H20 inventory charge rather than a sudden, permanent step‑change in economics. It flags China export controls and the tentative H200 licensing scheme as the biggest external overhang, notes 20.2 billion dollars of buybacks plus an 80 billion authorization and higher dividend, and calls out 13.4 billion of unrealised gains on equity securities as a big, less durable contributor to net income.
By the time the morning meeting starts, they have the numbers, the narrative behind them, and a clickable source link under every claim.
How to get started
Open OpenBB Workspace, go to to Apps → Apps Marketplace, select the AlphaCreek App and connect to the AlphaCreek MCP server.
Within seconds, you can start asking questions through OpenBB Copilot or any other agent you run in Workspace, and get sourced answers straight from SEC filings.
