Adaptive mode
Adaptive runs more than a single lookup. It can search, open, and read across your Knowledge in several steps before an answer is generated — and turn the result into a document you can download.
Adaptive uses the same Controls as default Chat. There is no separate Adaptive settings screen. Adaptive adds extra retrieval steps and Extensions, not a second behaviour UI.
Why it exists
A direct question needs one retrieval and one answer. A broad or multi-part question — a comparison, a summary of many sources, or "what do I have on this topic" — needs several passes.
Adaptive is for those questions. You ask once; extra retrieval steps run across your Knowledge before the answer is generated. Turn it on when the question is broad. Leave it off when you want the fastest answer to a direct question.
How it works
When Adaptive is on, retrieval runs in a short, bounded loop:
- Searches your Knowledge for relevant passages.
- Lists and opens the documents most relevant to the question.
- Reads a full document when a passage is not enough.
- Processes the retrieved material.
- Generates a grounded answer, with Sources attached.
Retrieval stops when a grounded answer can be generated, within a bounded number of steps. Every step runs inside your Space. Nothing leaves your environment, and your conversations are never used to train any model.
What you see
Each step appears in the chat as it runs. When the answer is ready, the steps collapse into a single line you can expand.
- Steps — labelled actions such as Knowledge search, List documents, Read document, Extension, and Find matching text. Expand Completed N steps to review them.
- Sources — the documents behind the answer, the same as any grounded answer.
- Response footprint — after the answer, a small panel shows the mode used (Fast, Adaptive, or Adaptive Extended), time worked, tool calls, and tokens.
Extensions and generated documents
Adaptive can turn an answer into a document you can download — a report, a spreadsheet, a formatted document, or a web page — built from your Knowledge.
You do not choose or configure these. The matching Extension is applied from the request. When a document is ready, a Document ready step appears with a download.
- Ask for a deliverable in plain language, such as "put this in a spreadsheet" or "draft a one-page summary".
- Download the file from the Document ready step.
- Downloads can expire. If a link no longer works, ask again to generate a new file.
Adaptive vs Deep search
Adaptive and Deep search solve different problems, and they work well together.
- Adaptive controls how many retrieval steps run — extra steps within the current scope.
- Deep search controls how wide retrieval searches — across all your My Spaces instead of only the current one.
Turn on both for a broad, multi-step question that could be answered from any of your Spaces.
Common questions
Your question was broad or had more than one part. Adaptive runs extra retrieval steps across your Knowledge before answering, so you do not have to break the question up yourself.
Yes. Every step runs inside your Space, and your conversations are never used to train any model.
The request may not have called for a deliverable, or generated documents may not be enabled for your Space. Ask for a specific output, such as a summary or spreadsheet.
Toggle it off in the chat actions menu. Chat returns to a fast, single-pass answer.
No. Adaptive uses the same Controls as default Chat. There is no separate Adaptive settings screen.
Only when Deep search is also on. On its own, Adaptive works within your current scope.