Open and Edit Deep Research Agent Notes Directly in AI Chat

AI Analyst Chat now opens any Deep Research Agent result as an editable note next to the conversation, and can distil a new note from a chat that's already underway. Conversation on the left, the note on the right. The chat handles the back-and-forth. The note stays clean.

If you've used split-panel views in tools like Claude or ChatGPT, you may know this pattern as "artifacts". The mechanics are similar. The content is what's specific to research: notes are grounded in filings, earnings calls, press releases, and management commentary, with citations carried through.

What the note is, and what the chat is for

A note is a structured document. A thesis paragraph, a peer-comparison table, a guidance ledger, the bullets you'd bring to a Monday meeting. It's the deliverable.

The chat is the messy half. Question and answer. Back-and-forth. Half-formed instructions. Dead-end side threads. None of that belongs in the document.

The two stay separate on purpose. Marvin reads what came out of the conversation and edits the note: a tightened thesis paragraph, a peer table that gets a new column, a risk section that picks up the guidance miss you flagged six messages ago. The conversation stays as it is. The note stays clean.

Two workflows

Open a Deep Research Agent result, then dig in through chat

Kick off a Deep Research Agent for an earnings preview, a primer, or a peer comparison. When it lands, the result opens as a note in the panel on the right.

That's the starting point. Now ask the questions you actually wanted to ask. "Why is the operating leverage assumption so high." "Pull capital allocation into its own section and weight it by what management said this quarter." "Drop the section on China demand. Replace it with a bull-bear table on margins." The note rewrites itself around the conversation.

This is the most common pattern we expect. The agent gives you a structured starting point. Chat lets you push on it without losing the structure.

Create a note from the chat you're already in

Most chats won't produce a note, but some get to a point where the answer is something you want to keep. Ask Marvin to turn the relevant parts of the conversation into a note, and the document opens in the panel on the right with what came out of the discussion. The chat continues. From there, the note keeps updating as you keep talking.

You don't have to decide upfront. The note can come into existence at the moment a thread is worth promoting.

Pick up an existing note

A note doesn't end with the conversation. Open one from a previous session, point Marvin at a new transcript or filing, and the document updates around what's already there. "Update the guidance section with what they said on the call this morning." "Re-do the peer comparison with the latest quarter."

A note you started in February doesn't have to be re-built in May. The work compounds across earnings cycles instead of getting redone.

What ends up in a note

Notes are markdown documents the agent maintains. Contents are whatever the analyst directs: a thesis summary, a peer-comparison table, a guidance ledger, a draft of the bullets you'll bring to a Monday meeting.

The text is grounded in the same primary content as the rest of the platform: filings, earnings calls, press releases, and management commentary. Citations carry into the note, so if a line is doing work the source it came from is one click away.

How this fits with the rest of the chat

Notes live inside AI Analyst Chat and work alongside what's already there: interactive charts and tables, slash commands and pop-out view, and the Deep Research Agents that produce the structured outputs notes are built to receive.

The shape of the workflow is what matters. Agents do the long-running production work. Chat does the iteration. The note is where the work that survives both lives.