Metabot AI

Metabot AI is an AI chat bot tool. Metabot AI brings natural language question answering, chart summaries, SQL generation, and semantic search into Metabase. Answers are inspectable, follow existing permissions, run against your own warehouse, and can use your own model provider key.

Metabot AI is the assistant layer inside Metabase, the open source business intelligence platform used by a large base of companies. It lets people ask questions about company data in plain language, receive charts and tables in return, and read an explanation of how each answer was assembled. The assistant builds its queries from the semantic layer and metadata defined in Data Studio rather than guessing at column names, so terms like annual recurring revenue resolve the same way for the AI as they do for the analytics team. The design priority is auditability. Every AI answer links back to the underlying query, which can be opened in a notebook editor or viewed as SQL and edited by hand. Access follows the permission rules already configured in Metabase, so a user cannot reach data through the assistant that they could not reach directly. Queries execute against the customer warehouse; Metabase does not ingest, store, or sync that data, which matters for teams with strict data boundaries or per-customer segregation. Beyond chat, Metabot offers one-click summaries of any chart or dashboard, AI-assisted SQL generation and debugging, and semantic search that matches dashboards and models by meaning rather than keyword. Organizations can bring their own model key (Anthropic, OpenAI, AWS Bedrock, Mistral, OpenRouter, Microsoft Azure and others) and pay the provider directly, or use the managed service on Metabase Cloud. Administrators set access per user group, cap usage by token or message count, apply system prompts, and review conversation logs in a usage dashboard. Metabot also reaches Slack, an official MCP server, and an Agent Workspace API for custom agents, with embeddable AI components for customer-facing products. This suits data teams already on Metabase, and companies that want AI exploration without a separate LLM vendor holding their data. It is less suited to teams wanting a standalone chat tool, or to organizations with no semantic model in place, since answer quality tracks the richness of the definitions provided.