Settings

All configuration lives under Settings, grouped in collapsible panels, taking effect immediately.

Settings: Language, Appearance, AI Gateway, Browser, Prompt, Data Retention

Settings: the AI Gateway panel (a set key is masked; leaving it blank keeps the current one)

Language

UI language (简体中文 / English), effective immediately after switching.

Appearance

Theme mode: Light / Dark / Follow System. Follow System adapts to the macOS light/dark appearance automatically.

AI gateway

AI generation and self-healing require an LLM, called through an OpenAI-compatible gateway:

Setting	Description
Gateway URL	OpenAI-compatible baseURL
API key	Masked once set; leaving it blank keeps the current key
Model name	Default model; shared by generation, self-healing and AI repair
Model has vision	Sends image attachments and screenshots directly to the model and enables the see observation tool; when off, no images are sent
Reasoning depth	Controls how strongly the model reasons; keep off for non-reasoning models
Agent max steps	Upper bound of the generation loop

Model choice

Prefer a model with vision — it works noticeably better: the AI can "look at" page screenshots to understand the layout, which markedly improves locating and operating complex pages and custom components; image attachments are also understood directly.

Browser

Setting	Description
Browser path	Empty = auto-detect system Chrome; or point to a browser executable

Viewport is fixed at 1920×1080; production uses system Chrome.

Prompt

Setting	Description
Split prompt	The prompt used to pre-split step plans; restorable to default

Data retention

Setting	Description
Generation log retention	Expired generation logs are cleaned up at startup

TIP

After changing gateway settings, run a simple case through generation first to confirm connectivity and model availability.