AI Assistant

Ask questions, generate, review, and run SQL, and use Query Tuning and Sample Data inside NeoSQL AI Chat.

Features

In AI Chat, handle general questions, DB metadata lookup, and SQL generation and execution in one conversation, and review AI explanations, Tool Activity, SQL, execution state, and results in the same flow.

NL2SQL

Turn natural-language requests into SQL for the selected DB context.

Query Tuning

Analyze SQL and estimated plans to suggest rewrites and indexes.

Sample Data

Generate test INSERT statements from a physical table's structure and constraints.

Chat-centered workflow: Run inline SQL cards without a separate AI-only editor, or open an editable copy in the regular SQL Editor.

Opening the AI panel, selecting context, generating SQL, and reviewing its execution result

Availability by environment

The current project type determines whether an account key or local key is used.

EnvironmentProjectsKey used
WebAccount projectsAccount key
Desktop (signed in)Account and local projectsAccount keys for account projects; local keys for local projects
Desktop (signed out)Local projectsLocal key stored on this device

Note: Signed-in Desktop can manage both key groups, but each request uses only the key that matches the current project type.

Basic workflow

Check the model and DB context for each request, then review the result.

1

Register a Provider Key

Register an account or local key for the current project type.

2

Open AI Chat

Open the AI panel from the project screen or press Ctrl/Cmd+I.

3

Select a model and context

Before sending, select the key and model and the Connections and Schemas allowed for this request.

4

Review SQL cards and results

Review the explanation, SQL, state, returned or affected rows, errors, and truncation.

DB access and execution controls

Each request combines its selected context with Connection-level policies.

  • Selected contexts are the strict allowlist of databases the AI may discover or access with Tools for that request.
  • With no context, general questions and copy-only SQL examples work, but DB lookup, executable SQL, and Open Editor do not.
  • AI availability, DML and DDL auto-apply, data transfer, and masking are Connection-level policies.
  • Database account permissions, Read-only and Confirm protection, and DDL permissions can require another confirmation or block execution.

See Project AI Settings for Connection policies.

Data sent to external providers

The following may be sent to the selected OpenAI, Anthropic, Google Gemini, or Ollama endpoint while producing an answer.

  • Your question, recent conversation, and feature-specific instructions
  • SQL generation requests or the SQL being tuned
  • Tables, views, columns, keys, indexes, relationships, and constraints in selected contexts
  • Actual DB values returned by approved read-only queries (up to 10 rows)
Caution:

Questions, SQL, and DB query results may be sent to an external AI Provider. Do not enter passwords, tokens, or personal data directly.

Looking for an external AI client?

External clients such as Claude Code, Codex, and Gemini use MCP settings and policies separate from the built-in AI Assistant.

View the NeoSQL MCP guide