Branching: Follow-Up Research

Branching lets you follow a thread of inquiry from any briefing without starting over. Each branch inherits the full context of its parent, making follow-up questions faster, cheaper, and more precise.

What Is Branching?

When you generate a briefing, the AI loads and synthesizes a set of source documents to answer your question. Branching lets you ask a follow-up question that builds directly on that same foundation. Instead of the AI re-searching the library from scratch, it already has the context loaded and can respond with that background in mind.

The result is a research session that deepens naturally — like interviewing an expert who remembers everything you've discussed so far.

Why Branching Is More Efficient

A top-level briefing question requires the AI to search the library, identify relevant documents, and load that context before it can even begin composing an answer. That retrieval work costs tokens. When you branch from an existing briefing, that context is already cached — the AI skips the retrieval step and works from what it already knows about your research thread. This makes follow-up branches noticeably cheaper in tokens than an equivalent fresh question.

Tip: If you know you want to explore a topic in depth, start with one broad anchoring question and then branch repeatedly from it. You'll spend fewer tokens than running multiple independent top-level queries on the same subject.

How to Branch

Generate a briefing as you normally would. Wait for the response to complete.

Click the Follow Up button (or branch icon) that appears below the completed briefing response.

Type your follow-up question in the input that appears. Ask about something specific that was mentioned, push back on a claim, or request elaboration on one part of the answer.

Submit. The AI responds in the context of the original briefing and all branches so far.

You can branch again from any response in the thread. There is no limit to how deep a branch tree can go.

The Branch Tree

Every briefing and its branches are organized into a tree in your briefing history. You can see the full structure — which questions led to which follow-ups — and navigate back to any point in the tree to read or branch from there. This makes it easy to maintain multiple lines of inquiry from a single starting point without losing your place.

Good Use Cases for Branching

Drill into a specific claim. The briefing says a program ran from 2008 to 2012. Branch to ask: "What specifically was happening in that program after 2010?"
Follow up on an entity chip. You clicked a blue entity chip and want to know more. Branch to ask: "What does the library say about this person's role specifically?"
Explore a contradiction. The briefing cited two sources that seem to conflict. Branch to ask the AI to address the discrepancy directly.
Get a different angle on the same evidence. The initial briefing covered what happened. Branch to ask about why, who else was involved, or what happened next.
Narrow from broad to specific. Start with a topic-level question to orient yourself, then branch into the specific aspects that matter most to your research.

Branching vs. a New Question

Use a new top-level question when you want to start fresh on a completely different subject. Use branching when you want to stay in the same research thread and go deeper. If you are not sure, branching is usually the better choice — it costs less and the AI's answers will be better calibrated to what you already know from the parent briefing.