Skip to content

Chats overview

Chats is the conversational workbench. It is where you talk to models, compare them side by side, reuse saved prompts, and — when you need more than a single model call — bind a full agent graph to the conversation.

The Chats pillar (at /dashboard) has two parts:

  • a sidebar listing your chat sessions, and
  • the chat area, where messages and the input box live for the selected session.

Each session holds one or more chats. A session keeps its conversation history so you can return to it later, and all sessions are scoped to your active workspace.

Use the ModelPicker to choose which model a chat runs on. Chatty supports several providers out of the box:

  • OpenAI
  • Gemini
  • Mistral
  • HuggingFace
  • Ollama (local models)

Provider keys are configured once in Settings. See Models & providers for the full picture.

You can run the same prompt against multiple models at once and compare their answers side by side. Add several chats to a session, give each a different model, and send your prompt to all of them. A ChatConfigPanel lets you tune per-chat settings.

Stop retyping your best prompts. The PromptPicker and PromptLibraryModal let you save prompts to a library and reuse them across chats. You can also configure external prompt search sources in Settings.

A chat does not have to be a plain model call. You can bind an agent graph to a chat, so the graph runs on every turn — complete with exposed parameters, persistent dashboard state, and live per-node progress shown in the ChatGraphProgress view.