Models

Every conversation has a model. The workspace default is used for new threads; you can switch the model on an existing thread without changing that default.

#Choosing a model

Use a deeper model for long contracts, multi-file questions, and anything that must hold a lot of context. Use a faster model for lookups, short drafts and triage. The picker in the thread lists what your plan currently includes.

Typical lineup (names change as providers ship):

• Claude Opus — long context, careful writing.
• GPT — general work and tool use.
• Gemini — long documents and native PDF reading.
• Open-weight options (GLM, Kimi, DeepSeek) when you want a different cost or latency point.

#Sub-agents

On longer jobs the assistant can run researcher and writer sub-agents underneath the main turn. They use the same workspace context and the same confirmation rules — they cannot send or write on their own.

#Data handling

Prompts sent to a provider follow the security retention rule: nothing is kept after the request, and nothing is used to train a model. The AI providers page lists the models and OpenRouter Zero Data Retention settings.