AI in DBCatz

AI in DBCatz is off until you turn it on, and DBCatz is complete without it. When it is on: Nothing is sent until you press an AI button; every one has Preview what is sent, and nothing it suggests runs without your click.

Turn AI on

Open Settings → AI (mod+,) and switch on Use AI in DBCatz. This switch belongs to DBCatz alone — turning AI on in another Catzoid app does not turn it on here.

The provider (shared with the Catzoid apps)

The Provider section carries the badge Shared with Catzoid apps: the provider, key and model you pick here are the same in DevCatz and DialCatz. See What the cats share.

Providers

Anthropic (Claude), OpenAI (GPT), Google (Gemini), Z.ai (GLM), DeepSeek, Mistral, xAI (Grok), Groq, OpenRouter (many models), Ollama (local, offline) and any OpenAI-compatible server — LM Studio, vLLM, LiteLLM or a gateway. A provider shows · key saved when it has a key.

API key

Paste the key and press Save; a Get a key link takes you to the provider’s page. A saved key shows Saved in the macOS KeychainWindows’ protected storage (DPAPI) (Catzoid AI), with Replace and Remove.

Address

The provider’s address is shown, and can be changed for Z.ai, Ollama and OpenAI-compatible servers.

• Z.ai — the notes under the field explain the Standard key, GLM Coding Plan keys and open.bigmodel.cn.
• Ollama — runs on your computer; pull a model first, for example ollama pull qwen2.5-coder.

Model

A searchable picker. Load models fetches the provider’s list (models that cannot chat are left out). Check … with the key sends one tiny request per model — up to 24 at a time — and marks each ✓ or ✗ with the reason.

Test sends one tiny request to check the key, address and model.

The feature switches

Each AI feature has its own switch, with a Sends: … line saying what it sends:

• Ask AI in the Query tab (mod+shift+K)
• Ask AI in Logs
• Ask DBCatz (mod+K)

A feature that is off hides its buttons, and the app’s main process refuses it too. These switches, and Send samples, are DBCatz’s own — they are not shared with the other apps.

Ask AI in the Query tab

Press Ask AI (mod+shift+K, Query → Ask AI…) in a Query tab and describe what you want — the ten customers who spent most last month.

• enter asks, shift+enter starts a new line.
• The answer streams in, with a live step (step n of m) and the seconds so far. Stop ends it.
• Each editor tab keeps its own thread: ask follow-ups and they show as You: chips. New thread starts over.
• The answer carries a badge — Reads, Changes data, Read-only connection: Run will refuse it or No query — an explanation, and chips for the tables it used.
• The query goes into the editor, formatted. It never runs by itself — read it and press Run. Every safety check applies as usual.

If a request takes more than 20 seconds, a hint suggests a faster model. Requests time out after 120 seconds.

When the feature is off, the button says Ask AI in the Query tab is off… with Open Settings.

What Ask AI sends

Preview what is sent shows the exact text (What Ask AI sends). It is the structure, never your rows:

• tables, columns, types, keys, indexes and estimated row counts;
• for a big schema (more than 60 tables): the table list first, then the structure of the tables the AI picked, plus one hop along their foreign keys;
• on Redis / Valkey: key-name patterns from a SCAN of up to 1000 keys, with ids turned into * — never values.

Send samples — a checkbox per connection — adds a few real values: 3 rows per table, each value cut to 60 characters (on Redis, one value per key pattern). It is off unless you tick it.

The schema is cached for 10 minutes, and dropped when you press ↻ or change the structure.

Fix with AI

When a statement fails, its result shows Fix with AI: Ask AI looks at the failed statement and suggests a fix. Like every suggestion, it does not run until you press Run.

Ask AI in Logs

In the Logs tab, ✦ Ask AI sits on each error row (and appears on hover on the others). The Logs bar also takes a question of your own.

• On a row, it explains that entry.
• Without a row, it explains the errors in view (up to 20) — or, if there are none, the 10 newest lines.

The answer has a summary, the cause, steps to fix it and, where it helps, a fixed query with Copy and Open in Query — Not run: check it in the Query tab, then press Run.

Before anything is sent, obvious secrets are masked: IDENTIFIED BY …, PASSWORD …, AUTH …, password='…', API keys and passwords in URLs. Preview shows What Ask AI in Logs sends.

Ask DBCatz

Ask DBCatz (mod+K, Query → Ask DBCatz…, or the floating ✦ button bottom right on every screen) turns plain words into steps. Since 1.0.1 the ✦ button keeps still while it is idle. If AI or this feature is off, mod+K opens Settings → AI instead.

It starts with suggestions such as Which table has an email column?, Show the newest orders, Back up this database, Switch to the dark theme and Which connections are on production?. New chat starts over and Preview shows what would be sent.

The answer is a list of steps, each with Run, Review…, Again, Retry or Open DevCatz as it fits. Nothing happens until you press one.

What Ask DBCatz can do

Destructive steps are red — truncate or drop a table, drop a database, flush keys. Review… explains what will happen, then the usual dialog asks you to type the name. A run query step that writes is reviewed the same way. On a Read-only connection, changes are dropped from the answer and refused again if they reach Run.

What Ask DBCatz sends

Connection names, engines, safety modes and whether each is connected; where you are in the app; the open connection’s databases; and the open database’s tables (with their columns when there are 40 or fewer) or its key patterns.

No hosts, users or passwords. Nothing connects to a database to answer.

Where the AI key is kept

The key sits in the macOS Keychain item Catzoid AI, one entry per provider,The key is encrypted with Windows’ protected storage (DPAPI) for your account, one file per provider in %APPDATA%\Catzoid\keys\, shared by the Catzoid apps. AI requests go straight from your computer to the provider you picked, and only when you press an AI button. See What the apps send.

Last updated 2026-11-01