Data AI Chatbot

Data AI Chatbot is a self-hostable data chatbot: you connect it to your APIs and databases through MCP (Model Context Protocol), chat in plain language, and open charts and documents in a side panel. Optionally, Proof-Carrying Numbers (PCN) show which numeric answers came from tool results versus the model alone.

Data360 Chat is the World Bank reference customization: it ships pre-configured for the Data360 MCP Server and World Bank development indicators—the setup this documentation describes. Other teams can reuse the same app with a different MCP server via MCP_SERVER_URL.

Maintained by the World Bank AI for Data — Data for AI team. On Data360 Chat, live indicators and metadata flow through Data360 MCP; PCN helps readers tell verified statistics from model guesses at a glance.

What you can do

Chat with data, not with spreadsheets

Ask in everyday language—for example “GDP per capita for Kenya and Tanzania since 2010” on Data360 Chat. The assistant calls MCP tools your server exposes (by default, Data360 tools to search indicators, load time series, and explain results) so users don’t memorize database codes.

See which numbers you can trust

Proof-Carrying Numbers (PCN) labels values in the reply. A verified badge means the number came from a tool result you can trace; unverified means the model answered without that grounding. No more guessing whether a statistic is “real.”

Charts, documents, and code—side by side

The artifact panel opens next to the chat for interactive charts (Vega / Vega-Lite), editable documents, spreadsheets, and syntax-highlighted code. Build a mini report while you talk.

Work the way your organization logs you in

Use guest try-out mode, email/password, Microsoft (MSAL), or (for Data360 portal embedding) Data360 sign-in—depending on how your deployment is configured—so the same stack fits pilots and enterprise.

Highlights at a glance


Streaming answers	Replies appear as they are generated; long answers can be paused, resumed, and (where enabled) you can follow extended reasoning steps.
Rich messages	Markdown, math (KaTeX), code highlighting, and image attachments (where allowed).
History & feedback	Past chats, vote on answers, and feedback so teams can improve the experience.
Multiple AI providers	Powered by LiteLLM—Azure OpenAI, OpenAI, Anthropic, Google, and more, depending on your setup.

Choose your path

I’m using the app

Start here

Getting started — Sign in (or try as guest) and send your first question
Chat features — Streaming, stop, regenerate, thinking display
Data analysis — Indicators, charts, and exploring development data
Documents & spreadsheets — Artifacts next to the conversation
File attachments — Images and limits
FAQ — Common questions

I’m building or extending it

Architecture hub — how the docs are organized
Architecture overview — scope and stack
System context — actors and external systems
Backend · Frontend
Authentication — modes and flows
Integrations — MCP, Data360, and related systems
Infrastructure — components and data flow
Local development — run the app locally
Development — tests and tooling (README)
DEVELOPER.md — contributor notes

I’m deploying or operating it

Deployment hub — deployment documentation index
Docker setup (detailed) — compose-oriented walkthrough
Docker — Docker Compose for development
Production — hosting and operations concerns
Environment variables — full reference
Troubleshooting — common issues
Runbooks — operational procedures
README — quick start — high-level clone-and-run path

I’m administering it

Admin guide — feedback, maintenance, users
Authentication (architecture) — sign-in modes and flows
Feedback review
Maintenance mode
User management

Security & API

Security hub — security documentation index
Security overview — auth, cookies, CSRF, rate limiting
Risk assessment — high-level security themes and practices
Live API docs (when the backend is running): Swagger at {BACKEND_URL}/docs and ReDoc at {BACKEND_URL}/redoc — see API

Open source

This project is open source (Apache-2.0 with the World Bank IGO Rider). Upstream credit to the Vercel AI Chatbot template is in the repository NOTICE file.

Questions or ideas? Use the repository issue tracker or contact the AI for Data — Data for AI team (see the main README on GitHub).