Skip to content

FAQ

Common questions about Data360 Chat and the Data AI Chatbot experience.

General

What is the difference between “Data AI Chatbot” and “Data360 Chat”?

Data AI Chatbot is the general application: connect data through MCP, optional PCN, artifacts, and multiple auth modes. Data360 Chat is the World Bank reference setup that uses the Data360 MCP Server and development indicators—the scenario this user guide describes.

Which AI models can I use?

Administrators connect models through LiteLLM (Azure OpenAI, OpenAI, Anthropic, Google, etc.). You only see what they enable. There may be a model picker in the UI, or a fixed default.

Is my history saved?

Signed-in users: chats are usually persisted in the application database. Guest users: persistence depends on guest session settings—you may lose access if the session expires or you clear cookies.

Can I delete a chat?

Yes, if your UI offers delete on the sidebar or chat menu. Deletion is typically permanent for that copy of the thread.

What does token usage mean?

If shown, it is an approximate count of tokens for the visible turn or session. Use it for orientation; exact billing or quotas are defined by your administrators and cloud provider.

Data, tools, and PCN

Why didn’t the assistant call Data360 for my question?

Common cases:

  • The model classified it as general knowledge (no tool needed).
  • MCP or Data360 was unreachable (see errors in the reply or try again).
  • Your phrasing was vague—try: “Search indicators for …” or “Plot … from Data360.”

What do PCN badges mean?

Proof-Carrying Numbers label whether a number in the reply aligns with a tool result. Verified means grounded in returned data; unverified means stated without that trace. See Data analysis.

The chart is blank or broken.

  1. Refresh once.
  2. Try another browser (extensions sometimes block chart scripts).
  3. If it persists, MCP, CDN, or CSP settings may block resources—contact your administrator with the time and chat id if safe to share.

Chat UI

Follow-up chips look wrong or noisy.

Chips are parsed suggestions from the assistant text. Ignore them or edit before sending. They are not official Data360 queries.

I don’t see thinking / tool steps.

Display depends on model, routing, and deployment. Some modes collapse reasoning. You should still see final answers and data.

Stop didn’t work instantly.

Stop cancels new tokens; a small delay is normal. If generation never stops, refresh and report the issue.

Files and uploads

Upload failed immediately.

  • Check size and format.
  • Try another browser.
  • Ask your admin for limits (MAX_UPLOAD_* / NEXT_PUBLIC_* settings).

Account and access

I was logged out suddenly.

  • Session expiry or password change.
  • Deployment invalidated cookies (maintenance, security rotation).
  • MSAL: parent portal may need re-login before embedded chat works.
  • Link revoked or expired.
  • Recipient lacks permission or must log in with an allowed account.

Getting help

  • Organization: internal support or admin for your deployment.
  • Open source / bugs: see the repository README for issue tracker and contacts.