Skip to content

Chat features

How the conversation behaves while you use Data360 Chat: streaming, tool and thinking display, stopping and retrying, feedback, follow-ups, and optional extras (tokens, math).

Streaming replies

Answers usually stream token by token instead of appearing all at once.

  • You can start reading immediately.
  • Long answers may take several seconds; data-heavy turns can take longer because the model may call tools first.

Tool use and “thinking”

For questions that need Data360 (or other MCP tools), the assistant may:

  1. Plan — Decide which tool to call (search, get data, chart spec, etc.).
  2. Show progress — You might see thinking or tool segments (wording depends on deployment and model).
  3. Answer — Combine tool results into a clear reply, sometimes with tables or charts inline.

If the answer feels slow

Wait for tool steps to finish. If the UI seems stuck, use Stop and shorten or narrow your question.

Stop generation

While the assistant is still generating:

  1. Click Stop (or the equivalent control).
  2. The partial reply stays in the thread; you can edit your last message or send a new one.

Regenerate

To ask for another answer to the same user message:

  1. Use Regenerate / Retry on the assistant message (exact label varies).
  2. The new answer may differ; it is not guaranteed to be “more correct”—rephrase if you need different data or constraints.

Resume after refresh (when enabled)

If resumable streams are enabled and your connection drops mid-reply:

  • Reload the page or return to the chat.
  • You may be offered resume behavior so the stream can continue or recover.

If nothing resumes, send the question again or use Regenerate if available.

Voting and feedback

Many deployments show thumbs up / thumbs down or a feedback control on assistant messages.

  • Use them for quality signals (helpful, inaccurate, unsafe wording, etc.).
  • They do not usually change the answer in real time; they inform operators and model tuning.

Follow-up suggestions

Some replies end with suggestion chips (short follow-up questions).

  • Click a chip to drop that text into the input (or send it—behavior depends on configuration).
  • They are generated from the assistant’s text; treat them as hints, not official Data360 queries.

Math, markdown, and code in replies

Replies often support:

  • GitHub-flavored Markdown (headings, lists, tables, links).
  • Code blocks with syntax highlighting.
  • Math via KaTeX when the model emits it.

Use normal copy from your browser to grab code or tables.

Token usage (when shown)

If your UI shows token or usage information, it reflects approximate consumption for that turn or session (definitions depend on your deployment). It is mainly useful for power users and cost awareness, not for precise billing unless your admin says otherwise.

Model selection

Some setups let you pick a model from the header or settings. Options are fixed by administrators; if you do not see a selector, the deployment uses configured defaults.

Maintenance mode

If the app shows a maintenance page, chat is temporarily unavailable. Try again later or contact your administrator.

See also