Choose a Model | Conversational Search

Switch between model families in Pieces and choose Fast, Balanced, or Extra Thinking, with optional Reflection Mode powered by the Agent Harness.

Model Selection

Click the model button in the bottom toolbar to choose your model. Pieces keeps selection simple: pick a model family, then choose a mode.

Pieces offers four model families:

  • Claude
  • Gemini
  • Grok
  • ChatGPT

Each family is available in three modes:

Mode Speed Quality Best For
Fast Fastest Good Quick questions, simple lookups, rapid iteration
Balanced Moderate Better General tasks, summaries, everyday coding help
Extra Thinking Slower Best Complex reasoning, debugging, multi-step analysis
Start with **Fast** for most tasks. Switch to **Extra Thinking** when you need deeper analysis. Click the `model button` in the *bottom toolbar*. Hover over **Claude**, **Gemini**, **Grok**, or **ChatGPT**. Click **Fast**, **Balanced**, or **Extra Thinking** for that family.

Your chat history stays intact when you switch models—new messages use the selected model while previous responses remain unchanged. Every family and mode is available on all plans.

Reflection Mode

Toggle the lotus icon next to the model selector to enable Reflection Mode. The agent reflects on its own reasoning and self-corrects in real time, producing higher-quality answers on harder questions.

Reflection Mode is powered by the Agent Harness—the framework that lets the agent reason across multiple turns, cross-reference your context, and refine its output before responding. For more on how this works in chat, see Using Conversational Search.

Web Search

Real-time web search with citations is powered by Perplexity. It runs automatically when a question needs current information; it is not a selectable model family.

Your Data & Model Choice

Want control over which specific models are used or where your data is processed? Bring your own keys (BYOK) through your organization. See BYOK & Org Models.


Next Steps

Learn how to filter your searches by Apps, Time, and Modality to scope which memories are used in your conversations.