AI Datavisor

TL;DR. AI Datavisor is a chat-style analytics console inside PanDev Metrics. It lives at /dashboard/datavisor/console — ask a question about your connected databases in plain language, and the AI writes the SQL, runs it read-only and presents the answer as text, a table or a chart. PanDev Metrics' own database is connected out of the box; you can add your own PostgreSQL and MySQL databases on top.

Console

The Console (/dashboard/datavisor/console) is where you talk to your data. Type a question into "Ask about your data…" and the AI works through three visible steps: Thinking → Querying → Summarising. The answer streams in as text, a table, or a chart — switchable between Bar, Line and Pie right on the message.

Conversations are organised into sessions in the sidebar, grouped by time (Starred / Today / Yesterday / This week / Earlier). A session can be renamed, starred, deleted; answers can be copied, downloaded as CSV, and the whole chat history exported.

Sources

Each session answers only from the sources you select. Use Add / Manage → All sources to pick which connections the chat sees. Sources are scoped to your department.

Connections

The Connections page (/dashboard/datavisor/connections) manages the databases Datavisor can query:

• PanDev Metrics is connected by default as a built-in source.
• Add your own databases — PostgreSQL and MySQL are available today; Microsoft SQL Server, Google BigQuery and others are coming next.

Adding a database is a three-step wizard: pick the type, fill in the credentials (host, port, database, user, password), and confirm — Datavisor checks access, verifies read-only privileges and snapshots the schema (takes a few seconds).

Each connection card shows its health: Healthy, Down, Schema drift (the database structure changed — refresh the schema), or Pending. From the card menu you can edit the connection, refresh the schema, disable/enable or delete it.

:::note Read-only by design Datavisor never modifies your data. Connections are verified for read-only privileges at setup, and every generated query runs read-only. :::

Skills

The Skills page (/dashboard/datavisor/skills) is where you teach the AI about your tables. The LLM only sees tables that have a description and a skill filled in — unconfigured tables are invisible to it.

For each table you provide:

• Description — one sentence: what is this table about?
• Skill — plain-text instructions for the LLM: which WHERE clauses to apply, which JOINs to make, how to count things correctly.

Filters (All / Configured / Without details) help track which tables are ready. When a connection's schema changes, the skill is flagged for an update.

AI provider

Datavisor needs a connected AI provider to work — Gemini, OpenAI, Anthropic, or any OpenAI-compatible endpoint (vLLM, Ollama, an internal proxy). If no provider is configured for your department, the console will prompt you to connect one (or to ask an administrator).

Access

AI Datavisor is available to owners and maintainers — at the tenant and department level. Connections, skills and chat sessions are scoped per department, so each department works with its own set of sources.