Short answer: DataGrip AI Chat needs schema context in plain language. Click the AI Chat input, query console, or Commit message box, press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux), speak table names, join shape, and output format, press again. AICHE inserts text. It does not run SQL or inspect schemas.
AI Chat, query console, and migration commits
DataGrip surfaces: AI Chat, query console, COMMENT ON TABLE, Commit dialog bodies, and EXPLAIN ANALYZE notes you paste above a saved query. SQL keywords stay short; the prompt and commit explanation are long.
Example AI Chat prompt: "Using schema analytics, join orders to customers on customer_id, filter created_at last 90 days, group by country, return top 10 by revenue descending, output as a SELECT only."
Example migration commit body: "Add nullable retry_count to webhook_events, backfill 0, index on (status, retry_count) for the dead-letter worker."
What Changes
Put the cursor in the field that needs the long text. Press the hotkey. Talk for as long as you need. Press the hotkey again. AICHE transcribes, removes the filler, and inserts the text in place. You keep your hands free for the parts DataGrip is already good at: pressing Run, accepting a completion, opening a Modify dialog.
Speaking runs about 150 WPM. Typing runs about 40 WPM. A 200-word query rationale that takes five minutes to type takes a little over a minute to speak.
How It Works
- Open DataGrip and connect to a data source.
- Click into the field you want text in - the AI Chat input, a
.sqlquery console, the Commit message box, a--comment line above a query, the table comment in the Modify dialog. - Press
⌃+⌥+R(Mac) orCtrl+Alt+R(Windows/Linux). Recording is a toggle, not push-to-talk. - Speak the prompt, comment, or message in full.
- Press the hotkey again. AICHE transcribes, cleans up, and inserts at the cursor.
AI Chat And AI Agents
DataGrip's biggest free-text surface is the AI Chat tool window once you install JetBrains AI Assistant. Newer DataGrip versions also expose Claude Agent and Codex inside that chat as AI Agents - a single multi-sentence prompt can drive multi-step edits across SQL files.
This is where voice pays off the most. AI Assistant is schema-aware: you can attach a table or a whole schema as context and ask for a query in natural language. The quality of what comes back tracks directly with how much detail you put into the prompt. Typing a 200-word prompt with the join shape, the filter window, the index you want it to use, and the format you want the result in is a chore. Speaking it isn't.
A real prompt sounds like a colleague briefing another colleague. The constraints, the table you already know is hot, the format you want the answer in. Talk it through, hit Enter, accept or refine the diff the agent proposes.
Query Comments And COMMENT ON
The block of -- or /* */ above a query is the only place future-you and on-call-you will look when the query starts misbehaving at 3am. The same goes for COMMENT ON TABLE and COMMENT ON COLUMN, which DataGrip exposes through the Modify dialog and which sit in the catalog where every other tool can see them.
Put the cursor on a blank line above the query (or in the comment field of the Modify dialog), press the hotkey, and talk through the reasoning: why this join order, why this index, what the cardinality looks like in production, what the failure mode is when the input is empty. AICHE inserts the prose. Add the -- markers and you have a comment that explains the query instead of restating it.
Same workflow for the description on a long view, a stored function, or a partitioned table. Speak the trade-off once, paste it into the catalog, stop re-explaining it in standup.
Commit Messages For Migrations
DataGrip ships the JetBrains Commit tool window. The message box is the place to write what most teams instead bury in Slack: which tables changed, whether the migration is reversible, what to do if it fails halfway, what window to deploy it in.
Click into the Commit message field, press the hotkey, talk through the migration. The result is a paragraph reviewers can actually use, not a three-word title that means nothing in git log six months later. Same trick for branch descriptions and push tag annotations.
Code-Heavy Speech
DataGrip work is full of identifier names, kebab-case CLI flags from psql or mysqldump, snake_case column names, and product names that fight a generic transcriber. Turn on the Software Development profile in AICHE (Pro). It is tuned for code-shaped speech and respects names like pg_stat_statements, JSONB, EXPLAIN ANALYZE, and ON DELETE CASCADE instead of trying to translate them into English.
Drop your repo names, internal service names, and table prefixes into Custom Vocabulary so they come through spelled correctly. Turn smart quotes off so a dictated quote does not break a psql paste.
Multilingual Thinkers
Turn on Auto-translation in AICHE settings if you think in a language other than English. Talk through the query rationale in German, Mandarin, Spanish, Japanese, whatever your brain runs on. AICHE outputs English text in the editor or chat. AI Assistant reads clean English. You stop translating in your head while typing.
Common Questions
Q: Does AICHE need a DataGrip plugin?
A: No. AICHE runs as a desktop app and inserts at the system cursor. DataGrip sees the text the same way it sees a paste from your clipboard. No plugin, no IDE settings change.
Q: Does it work in the AI Chat input specifically?
A: Yes. The AI Chat input is a normal text field. Put the cursor there, press the hotkey, speak the prompt, press the hotkey again, hit Enter. Same for chat with Claude Agent or Codex.
Q: Will it mangle SQL keywords and identifiers?
A: Use the Software Development profile (Pro). It is tuned for code-shaped speech - SQL keywords, snake_case, kebab-case flags, library and engine names. Add your schema names to Custom Vocabulary and they come through correctly every time.
Q: I'm on Linux with Wayland. Does the global hotkey work?
A: Usually yes, but some Wayland environments need extra accessibility/keyboard permissions or compositor-specific settings. If it does not trigger immediately, check the Linux setup notes and permission prompts first.
Q: Can I dictate into the query console while a long-running query is still streaming results?
A: Yes. The query console is a normal editor. AICHE inserts at whatever editor has focus.
Q: Does this replace the AI Assistant plugin?
A: No. AI Assistant generates the SQL. AICHE gets your words into AI Assistant. They sit on different sides of the same problem.
Q: My commit messages end up in git log. Does AICHE keep my audio?
A: Audio is streamed for cloud transcription, processed, and discarded immediately after processing, within 1 second. No persistent audio copy. See the privacy page.
Result: the SQL comment above the migration, the AI Chat prompt that produces the right join on the first try, and the commit message reviewers can read in six months all get written instead of skipped. DataGrip stays fast. The keyboard stops being the slow part.
Try it now: open DataGrip, attach a table as context in the AI Chat, press your hotkey, and spend sixty seconds describing the report you actually want - the joins, the filters, the format. Hit Enter and see what the agent comes back with.