AI providers¶
Each provider keeps its own key, model and base URL. Test connection loads the live model list.
| Provider | Default model | Chat | Embeddings |
|---|---|---|---|
| OpenAI | gpt-4o-mini |
yes | yes |
| Anthropic | claude-haiku-5-5 |
yes | no |
| Google Gemini | gemini-3.6-flash |
yes | yes |
| DeepSeek | deepseek-chat |
yes | no |
| OpenRouter | openai/gpt-4o-mini |
yes | no |
| Cohere | command-r7b-12-2024 |
yes | yes |
| Custom (OpenAI-compatible) | your choice | yes | yes |
| Ollama (local) | llama3.2 |
yes | yes |
Custom works with LM Studio, vLLM, Groq and similar; the key is optional. Ollama uses http://127.0.0.1:11434/v1 and needs no key.
Keys are stored in Zotero prefs and sent only to the provider's base URL.