Skip to main content
Babbily can use search tools when a question needs current information, source links, or broader research. Use web search for focused questions that need current sources. Good examples:
  • “Search the web for the latest pricing changes from this vendor.”
  • “Find recent articles about this market and summarize the pattern.”
  • “Compare these two tools using current public information.”
  • “Show me photos of the new stadium and summarize the latest construction updates.”
Search replies can include citations so you can review the sources. Search can also add a few pictures above the answer when the subject is something you can see or when you ask to see it. For the first question in a chat, Babbily decides whether pictures help. In follow-up questions, Babbily adds search pictures only when you ask for them, and it leaves out logos, icons, ads, and pictures that do not match your question.

Visual replies in chat

Search replies may include citations, pictures of a visual subject, charts, or other visual formats when they help. Learn how to ask for pictures from web search.

​
Deep Research Agent

Use Deep research when you need more than a quick lookup. The Deep Research Agent plans a broad question, works through focused sub-questions, gathers sources, and writes a synthesized report into Canvas. The report cites text sources and does not include pictures from web search. Good examples:
  • “Research the competitive landscape for AI meeting assistants.”
  • “Build a sourced brief on recent changes in enterprise AI procurement.”
  • “Find the strongest arguments for and against this product direction.”
Deep research can take several minutes and may use more API usage budget than a normal reply. Use web search when one focused lookup is enough.

​
What happens during a run

1

Babbily plans the research

The agent turns your request into a concise title, research questions, and a report outline. If the scope is unclear, Babbily may ask you to clarify it first.
2

Canvas opens with the plan

The report opens in Canvas before research begins. You can follow the questions being answered, the planned sections, progress, and sources found so far.
3

The agent gathers and checks evidence

Babbily works through the plan with available research tools and reads from Plugins you selected. Source availability, blocked pages, and account permissions can affect coverage.
4

The cited report streams into Canvas

Babbily writes the report into Canvas and returns a short summary in chat. Review the full report and its sources before relying on important conclusions.

​
Turn search or research on

1

Open the composer menu

Use the tool control beside the message box.
2

Choose Auto or Manual

Use Auto tool mode if you want Babbily to decide whether a tool fits. Use Manual > Deep research when you want to run the agent explicitly. Explicit Deep research needs a named model other than Claude Opus 5.5 or Claude Sonnet 5.5; with Auto selected in the model picker, use Auto tool mode and ask for a research report instead.
3

Ask a source-aware question

Include the scope, entities, timeframe, sources that matter, and how you want the report structured.
Tool mode is separate from the model picker. Choosing Auto in the model picker lets Babbily choose the model; Auto tool mode lets Babbily choose tools.

​
Use Plugins during research

When you select Plugins for a run, Deep Research reads from them only for steps that read data, such as finding a document or an email thread. It doesn’t send, create, or update anything in a connected app, and it doesn’t pause to ask for approval. If you need an action in a connected app, such as sending an email or updating a document, ask for it in a regular message after the report is ready. Babbily then shows an approval card for that action. See Plugins. In a chat that uses shared project instructions, Deep Research may skip Plugin reads.
A Deep Research report can include information from connected apps you chose to use. Review the report before you export it or share the chat.

​
When a report ends early

If evidence gathering reaches its time limit, Babbily finishes a report from the available evidence and tells you its coverage is partial. If report writing reaches its time limit, Canvas keeps the written portion and labels the report incomplete. Review what it covers, then narrow the scope and run a follow-up when you need the missing sections. If the run fails before it can produce a report, try a narrower request, remove an unavailable source or plugin, switch to a supported model, or retry later.

​
Read citations carefully

Citations help you check where an answer came from. They are especially important for fast-changing topics, pricing, policies, legal details, medical topics, and investment research.
Search and research can summarize public or authorized connected-app information, but you should verify high-stakes decisions directly with authoritative sources.

Built-in tools reference

Compare web search, deep research, site map, finance research, and other built-in tools.

Research recipe

Follow a step-by-step research workflow with citations.

Finance

Use Finance when your research is about markets, companies, earnings, prediction markets, or watchlists.
Last modified on October 9, 2026