AI in DevCatz is optional and off until you switch it on. With it on, DevCatz can explain a crashed site and offer one-click fixes, draft your commit messages, explain a webhook, and turn plain words like start shop and its database into actions — always waiting for your click before anything runs.
You bring your own provider and key. DevCatz talks to that provider directly from your computer; nothing goes through catzoid.com.
Settings → AI
| Option | What it does |
|---|---|
| Use AI in DevCatz | The main switch. Off by default. |
| Provider | Which AI service to use (below) |
| API key | Replace, Remove, or Get a key to open the provider’s key page |
| Address | For providers with a custom server address (Z.ai, Ollama, an OpenAI-compatible server) |
| Model | Pick one, or Load models to ask the provider for its list; Check … with the key marks the models your key may use with ✓ / ✗ |
| Test | Sends one tiny request to see that everything works |
| Features | A switch per feature. The AI buttons hide for a feature that is off. |
The provider, the model and the key are shared with the other Catzoid apps: set them once in DevCatz, DBCatz or DialCatz and the others follow. The on / off switch and the feature switches are not shared — each app starts with AI off. See One AI connection.
The key is kept only on this computer — in the macOS Keychain, as the item Catzoid AIin Windows’ protected storage (DPAPI), for your account only — never shown again and never sent anywhere but your provider. Settings → AI says so under the key: Saved in the macOS KeychainWindows’ protected storage (DPAPI), shared with Catzoid apps.
Providers
| Provider | Notes |
|---|---|
| Anthropic (Claude) | The default |
| OpenAI (GPT) | |
| Google (Gemini) | |
| Z.ai (GLM) | Use a standard pay-as-you-go API key; keys from bigmodel.cn use the https://open.bigmodel.cn/api/paas/v4 address |
| DeepSeek | |
| Mistral | |
| xAI (Grok) | |
| Groq | |
| OpenRouter | |
| Ollama | Runs on this MacPC, nothing leaves it — install it from ollama.com, then for example ollama pull qwen2.5-coder |
| OpenAI-compatible server | Any server with an OpenAI-style /chat/completions: LM Studio, vLLM, LiteLLM, a company gateway |
The features and what each one sends
Each feature sends only what it needs, with secrets masked first.
| Feature | Where | What it does | What it sends |
|---|---|---|---|
| Explain & fix errors | ✦ Explain with AI on a crashed site (Overview) or on lines in Logs | What went wrong, with one-click DevCatz fixes. You can ask a follow-up question. | The log lines (secrets masked), the stack and runtime versions, the dev command |
| Commit messages and PR descriptions | ✦ Write with AI in the site’s Git tab | A commit message from the staged changes, or a pull request description, in the repository’s own style | The staged diff (for a PR: the branch’s commits and diff) and recent commit subjects |
| Docker | ✦ Explain the last Docker output and ✦ Edit with AI in the Docker tab | Why a build or docker compose up failed; changes a generated Docker file from plain words |
The compose / build log, and the file you adjust |
| .env helper | ✦ Check with AI in the site’s Env tab | Explains empty or odd keys against .env.example and suggests values pointing at DevCatz services |
Key names, whether each is set, .env.example values, DevCatz service addresses — never your .env values |
| Search playground | ✦ Ask AI in the search playground | Plain words become the engine’s query, filter and sort | The index’s settings (fields, filterable / sortable attributes) and your request |
| Webhooks (Expose) | ✦ Handle with AI in the Expose request inspector | Explains a request that came in through a share and writes a handler for the site’s framework | That request’s method, path, headers (auth masked) and body |
| Ask DevCatz (mod+K) | Anywhere, with mod+K | Plain-word commands turned into DevCatz actions you confirm | Site and service names, domains, status and stack — no code, no .env |
Code diffs are sent only when you press the commit button, never in the background.
Ask DevCatz
Press mod+K anywhere to open Ask DevCatz, a chat palette. Say what you want in plain words — restart the shop site, switch blog to PHP 8.3, share api with a public URL — and DevCatz works out the steps.
Nothing runs until you press Run. Each step is shown as a button first, so you can see what will happen.
- enter sends, shift+enter adds a new line, esc closes.
- New chat starts over.
- Preview shows exactly what would be sent, without sending it.
mod+K only works while AI and the Ask DevCatz feature are both on.
What Ask DevCatz can do
| Kind | Examples |
|---|---|
| Sites | Start, stop or restart a site; install its dependencies |
| Runtimes | Set a site’s Node.js, PHP or Python version; turn on a PHP extension |
| Tasks | Run a database task or a stack task (migrations, seeds and the like) |
| Services | Start, stop, restart or uninstall a database, storage, WebSocket, search or gateway service |
| Expose | Start or stop sharing a site |
| Site settings | Start automatically, queue worker, scheduler, HTTPS, linked services |
.env |
Set a value |
| Navigation | Open a site tab or a section; open a database in DBCatz |
Privacy
Secrets are masked
Before anything is sent, DevCatz masks what looks like a secret:
KEY=secretvalues,- bearer tokens,
- keys starting with
sk-,ghp_,glpatorAKIA, - hex strings of 32 characters or more,
- passwords in URLs,
- JWTs.
Your home folder shows as ~. Masked values appear as •••.
Preview what is sent
Preview what is sent (in Settings → AI, and Preview in Ask DevCatz) shows the exact request: the instructions with what DevCatz knows (system), then the conversation. Exactly that, nothing more.
Suggestions are buttons
The AI never changes anything by itself. Fixes and actions come back as buttons, and you decide which ones to press.
More on what the apps send, AI included: What the apps send.
Good to know
- Your provider’s own terms and prices apply to what you send it. With Ollama, nothing leaves your computer.
- Answers can be wrong. Read a suggested fix before you press it — the cat is clever, not infallible.
Last updated 2026-11-01