Settings: AI models
Providers, default models for chat, titles, translation, OCR and compression, prompts and retries.
Settings → AI models has four entries.
Providers#
"CX AI, plus any provider you add with your own key". See CX AI models and Bring your own API key. Each provider page has Configuration (enable/disable, keys, URLs, options, Test Connection, Disconnect/Reset, Share, delete) and Models. A balance option can read an account balance from an API path for supported providers.
Default models#
| Setting | Purpose |
|---|---|
| Chat Model | Global default chat model (also the fallback model). |
| Fast Model | Titles, chat suggestions and other fast or background tasks. |
| Translation Model | Message, page and Translator translations. |
| OCR Model | Text recognition on images. |
| Compress Model | Compressing conversation history. |
| Response stream retries | Retries when a Response API stream closes early (0 disables). |
| Enable automatic context compaction | Plus threshold (percentage or token count), keep recent tool calls, compression target. See Long chats. |
| Thinking Budget | Per-model reasoning budget where supported. |
Prompts#
Edit the built-in prompts with Reset to default:
- Translation – variables
{source_text},{target_lang}. - Title Generation –
{content},{locale}. - OCR Recognition –
{images}. - Context Compression –
{content},{target_tokens},{additional_context}.
MCP servers and Orbits#
See MCP servers and Orbits.