Docs/AI Chat

AI Chat

AI Chat is the AI assistant built into the DBConvert Streams desktop app. Open it, pick an agent, and ask about your databases, files, and streams in plain language. It works from your live workspace — the same connections, schemas, files, and stream status you see in Data Explorer — and it is read-only: it can look and advise, but never changes your data.

It runs your own agent (Claude Code, Codex, GitHub Copilot CLI, or OpenCode) with your own model key — DBConvert injects its tools into the session automatically, so there's no separate setup.

Want to use DBConvert from an external AI app (Claude Desktop, Cursor) or over HTTP instead? See AI Clients (MCP).

Start a chat

  1. Open the DBConvert Streams desktop app.
  2. Click AI Chat in the left sidebar.
  3. Pick an installed agent, and start a session.
  4. Ask: List my DBConvert connections. — if the connections you see in Data Explorer come back, you're ready.

No agent installed? AI Chat runs an agent you already have on your machine. Install one of the supported agent CLIs and it appears in the picker.

Ask AI in context

You don't have to start from a blank chat. Ask AI buttons across the app open a chat already scoped to what you're looking at, so the assistant answers about that thing without enumerating your whole workspace first:

  • a table or view (Data / Structure / Summary),
  • a file (CSV / JSON / Parquet),
  • a connection or database,
  • a SQL console (including a multi-source federated one),
  • a stream config (from its Monitor / Compare / History tabs).

Each scope also offers a few one-click suggestions tuned to it — e.g. "Find data quality issues in this table", "Show the first rows" for a file, or, in a multi-source console, "Compare a table with a file".

What you can ask

Real prompts, real answers — every exchange below came back from a live workspace, numbers and all.

Database inspection

You: Describe the orders table on my Postgres demo connection — columns, keys, indexes. Lay it out as a table.

AI: public.orders — 10 columns, primary key order_id … No index on region — add one if you group orders by region.

Analytics — real numbers, not a guess

You: What are my top film categories by rental revenue? (sakila sample)

AI:

CategoryRevenue
Sports$5,314.21
Sci-Fi$4,756.98
Animation$4,656.30

Joined category → film_category → film → inventory → rental → payment, ran in ~55 ms. Want the query to reuse?

Stream troubleshooting

You: My MySQL → PostgreSQL CDC stream stopped overnight. List my streams, check status and recent errors, and tell me what to fix.

AI: Found it — the latest run of 10M-cdc-mysql2pg FAILED with Unknown database 'target', source-side. The CDC source points at a database that doesn't exist on the MySQL server. Fix: create it (or correct the name on the connection), then restart.

More complete transcripts: plain-English SQL exploration and federated Postgres + Parquet validation.

Files and multi-source

AI Chat isn't limited to databases.

  • Files. For a CSV / JSON / Parquet file, the chat reads its schema, previews rows, and queries it via DuckDB — no hand-written read_csv.
  • Multi-source (federated). In a SQL console with two or more sources attached — say a MySQL database and a folder of CSVs — AI Chat sees every source and its alias. Ask it to join, find the join keys, or compare and reconcile across them, and it writes the cross-source query and runs it. This is the everyday "is my CSV export still in sync with the live table?" check, done in one question.

Generate SQL

Next to any SQL console you'll also find Generate SQL: type a plain-language request and it writes the SQL straight into the editor (including federated queries across multiple sources), for you to review and run. It's the same engine as AI Chat, aimed at the editor instead of a conversation.

Privacy & safety

  • Read-only. Only SELECT / WITH … SELECT reach your database; everything else is rejected before it runs. No INSERT / UPDATE / DELETE, no DROP / ALTER / CREATE, no connection or stream changes. When the AI suggests a change, you run it.
  • Your keys, your model. The agent uses your own account; your model key never passes through DBConvert.
  • Credentials never leave. Tools return schemas, rows, and stream state — never passwords or cloud keys. Tool results and sample data do reach your model provider (that's how any AI assistant works); nothing is sent anywhere on its own.

Full boundary: Safety & privacy.

PreviousDatabase DiagramsNextOverview