AI agent
AI Agent
The Sheeta Agent answers questions about your databases in plain language. It reads your schema, runs read-only queries, explains the results and can build dashboards for you. The agent runs on Claude Code or Codex installed on your computer. If you have not set one up yet, start with Agent Setup.
Open the agent panel
Press Cmd+I (macOS) or Ctrl+I (Windows), or click the AI Assistant button at the right end of the tab bar. The panel opens in the right sidebar. A project must be open, because chats are saved in the project folder.
What the agent can do
The agent works through a fixed set of Sheeta tools:
- List your connections, without any credentials.
- List tables in a connection and describe their columns.
- Run read-only
SELECTqueries. - Read the AI permission rules saved in the project.
- List dashboards and create, update or delete charts, slicers and text on a dashboard.
The agent cannot insert, update or delete data, or change your schema. Any query that is not a SELECT is rejected. It only changes dashboards when you ask it to.
Examples of questions you can ask:
- "Show me all orders from last month"
- "What does this table store?"
- "Why is this query failing?"
- "How are users and orders connected?"
- "Draw me a line chart in a new dashboard to show sales trends."
Context the agent receives
With each message, Sheeta AI tells the agent which connection and schema are active, what kind of tab you are looking at, and which dashboards exist. The agent then looks up tables and columns itself with its tools. Earlier messages in the same chat tab are included, so you can refine a request step by step.
Working with results
Each tool call appears in the chat as a card, such as Query:, Table Info:, Dashboard: or Chart:. Click a dashboard or chart card to open that dashboard in a tab. Use the copy button under a reply to copy its text.

Attachments
Click + in the chat input to attach files, or paste an image from the clipboard.
- Images: PNG, JPG, WebP, GIF, BMP and SVG. The selected model must support images. Large images are resized before sending.
- Text files:
.sql,.md,.jsonand.env. Each file can be up to 256 KB, with up to 10 files and 512 KB in total per message.
Chat tabs and history
The agent panel has its own tabs, so you can keep several conversations open. Click + or press Cmd+T / Ctrl+T while the panel is focused to start a new chat. Open the More options menu and choose Chat History, or press Cmd+Y / Ctrl+Y, to reopen earlier chats.
Chats are stored in the project under .sheeta/ai-chat/. Each project has its own history.
Stop or report a response
While the agent is working, the send button turns into a Cancel button that stops the CLI. To report a harmful or wrong answer, click Report this response under the reply. Reporting requires signing in to a Sheeta account in Settings > Account.
AI permissions
Open More options > AI Permissions, or press Cmd+/ / Ctrl+/, to manage permission rules for the project. Rules are grouped into Allow, Deny and Ask, and each rule targets an action (such as Select, Insert or Drop) on a connection, schema, table and column.

Rules are saved in .sheeta/ai-permissions.json in the project. The agent can read them, and the sheeta query command enforces them (see CLI and MCP). In Sheeta AI 0.5.2, the in-app agent does not check these rules before running a query. Its read-only tool set is what keeps your data unchanged. The agent can query any connection in the project, so keep connections with data you do not want to share with a model provider in a separate project.
Privacy
Queries run on your computer. Database credentials are never given to the agent. Query results that the agent reads are sent to the model provider of the CLI you use (Anthropic for Claude Code, OpenAI for Codex), under that provider's terms.