Marvin 2.0 Is Live in AI Analyst Chat and Deep Research Agents
Marvin 2.0 is live. It is the AI behind AI Analyst Chat and Deep Research Agents.
The highlights:
- Native support for multi-company analysis
- Faster, more accurate financial lookups
- Improved citation fidelity for text and numbers
- File attachments in AI Analyst Chat
- More than 30 optional data sources through the connector catalog
Preliminary internal and external benchmarks put Marvin 2.0 ahead of the alternatives we tested. We will publish the full results in the next few days.
Native support for multi-company analysis
Compare the most recent annual gross margin for Apple, Microsoft, and SAP, and you get one table.
Until now the AI was bound to one company per thread, a constraint we kept to protect the quality of in-depth research. It could look across to another name when it needed to, but the thread belonged to the company it started on, and that made for some awkward exchanges.
Marvin 2.0 removes the constraint. A company analyst still owns the in-depth research on one name. Above it sits a coordinating agent that resolves the companies in your question against our coverage and delegates to one analyst per company, so a comparison across three names comes back as one answer.
Faster, more accurate financial lookups
Financial lookups sit at the center of any AI for investors, so this release carries a series of small improvements to them. Lookups now run in parallel: ask for revenue and EBITDA for the last four quarters and each metric gets its own lookup. They distinguish close but different metrics, such as adjusted and as-reported margins. And they understand composite periods such as LTM, nine months, and first half without being told how to build them.
The result is correct financials, returned faster, each with its source citation.
Improved citation fidelity for text and numbers
Text search had the same treatment. Finding the right passage is faster and more precise, using agentic search over the document text. One ranked list now spans the filings, press releases, and transcripts ingested by Automated Data Import, AI-generated results, and any internal documents you upload to Marvin Labs.
Answers cite each document as a numbered link. Hover one to see the document and its publication date, and every answer ends with a list of its sources.
File attachments in AI Analyst Chat
You can attach a file to a chat message from the composer, with the paperclip or by dropping it in. Until now the route was to upload it as an internal document first. That step is gone, and the list of formats is longer: PDF, text, Markdown, CSV, and images, with more to come.
The file goes to the model through its own file handling, so there is no parsing step before you can ask about it. After the answer, the conversation asks what you want done with the file. Keep it and it is copied into your documents library with the same parsing and indexing as a private document upload. Do nothing and it expires after 24 hours. Nothing joins the library without an explicit yes.
More than 30 connectors
Connectors are how Chat and agents reach data Marvin Labs does not already ingest. The catalog lists more than 30 sources. Six we host and switch on for you: US federal legislation from GovInfo, FRED macro series, Bundesbank yield curves, EIA energy statistics, Ember electricity data, and CFTC futures positioning. The rest are third-party sources you connect with your own account, through a sign-in or one pasted key, among them Daloopa, Fiscal.ai, FactSet, LSEG, S&P Global, Morningstar, and Interactive Brokers. Each one has a detail page saying what it covers and what connecting asks of you.
Common questions about Marvin 2.0
Alex is the co-founder and CEO of Marvin Labs. Prior to that, he spent five years in credit structuring and investments at Credit Suisse. He also spent six years as co-founder and CTO at TNX Logistics, which exited via a trade sale. In addition, Alex spent three years in special-situation investments at SIG-i Capital.
