The Analysis Chat

Every analysis has a dedicated chat with an AI research assistant. The chat is aware of the analysis topic, the research Hinsley has gathered, and the decompositions and scenarios you have created. It is designed to help you think through the topic, take structured actions on your behalf, and cite source material as it reasons.

Where the Chat Lives

The chat appears in two places:

• On the analysis home page - The full conversation occupies the left pane of the analysis page, with the research results tabs on the right.
• Throughout the analysis - A chat button is available on most analysis sub-pages (decomposition builder, scenario sets, forecasting questions, and others). Opening it brings up the same ongoing conversation in an overlay, so you never lose your place in the page you were working on.

Each analysis has a single, continuous conversation. The history is preserved across sessions and browser reloads - you do not start a new chat each visit.

When the Chat Is Available

The chat is disabled until the initial research workflow finishes for the analysis. While initial research is running, the input is locked and a message explains that the chat will become available once research completes. This ensures the assistant always has the research brief and source material to ground its responses.

What the Chat Knows

The chat has access to the following context automatically:

• The analysis topic
• The current research brief and any update summaries
• The active decomposition (drivers and indicators) and active scenario set, when those exist
• All source material documents saved to the analysis - including uploaded files and saved URLs - which the assistant actively searches and reads when composing responses

When you open the chat from a specific forecasting question, the question text, background information, resolution criteria, and any documents attached to the question are also available for that exchange. If the analysis has source material, the assistant can additionally search the broader analysis document library - not just the documents linked directly to the question - giving it access to the full body of curated evidence when reasoning about probability and resolution.

What the Chat Can Do

Beyond answering questions, the assistant can take structured actions directly in your analysis using a set of tools. Depending on context, these include:

• Search and read source material - The assistant can query the analysis’s research results and pull in content from specific documents, citing them in its replies.
• Generate or modify a decomposition - Generate a full decomposition, add new drivers or indicators, or replace existing ones based on the conversation.
• Generate or modify scenarios - Create a first set of scenarios, or add and replace scenarios in an existing set.
• Search the web - For some models, the assistant can run fresh web searches when the existing research does not cover a topic.

When the assistant makes changes - for example, adding drivers to a decomposition - the change is applied immediately to the relevant record. You can review and edit the result from the decomposition or scenario page as you would any other content.

Switching Context

If an analysis has multiple decompositions or multiple scenario sets, a context selector appears at the top of the chat. Use it to tell the assistant which decomposition or scenario set its actions should target. When only one of each exists, no selector is shown and the assistant works on the single active one.

Citations

When the assistant uses source material to support a claim, it cites the underlying document inline. Hinsley renders these citations as links you can click to review the source.

How It Differs From Other AI Features

The analysis chat is the most general AI surface in Hinsley, but several other AI features act independently of the chat:

• AI forecasting generates forecasts for individual questions on a schedule and produces its own rationale - it does not run through the chat.
• Question autofill on the Creating Forecasting Questions form drafts a question, background, and resolution criteria from a short prompt.
• Decomposition generation run from the decomposition settings menu produces a full decomposition with different prompts than the chat uses.

You can reach similar outcomes through the chat, but these dedicated features give you finer control over the specific task and its inputs.