Chats & Projects

Chatting with one model or several, Auto, web search and reasoning, attachments, voice, and keeping chats in projects.

A chat on Perspective AI can talk to one model or several at once. You choose them, or let Auto choose for you, and you can change your mind at any point in the conversation.

Choosing models

Start with New Chat. The models above the message box answer your message:

  • One model — pick it and chat as you would anywhere else.
  • Several models — add more and each one answers the same message side by side, so you can compare them or take the best answer.
  • Auto — switch the toggle beside the models to Auto and Perspective AI picks a suitable model for each message. Switch to Pick to choose yourself, and Pick again to open the full list.

The full catalog, and how models are added and priced, is on the Models page.

Web search and reasoning

The settings button beside the message box has two controls, offered for models that support them:

  • Web search — lets the model look things up before it answers.
  • Reasoning — how hard a reasoning model thinks first: Off, Low, Medium or High. More thinking takes longer and uses more credits.

Files, photos and voice

Use + to attach photos and files (PDFs, documents and more) for models that can read them. The microphone records a voice message and turns it into text in the message box, ready to edit and send.

Image and video models answer in chat too: describe what you want and the picture or clip arrives as the reply. For more control over size, length and style, use Studio.

Projects

Projects keep related chats together in the sidebar.

  • New Project in the sidebar creates one; give it a name.
  • Start a chat inside a project, or move an existing one in from its menu (⋯ → Move to project).
  • Rename or delete a project from its own menu. Each chat's menu also has Rename chat and Delete chat.

Other kinds of chat

  • Private Mode — the Private button at the top of the sidebar keeps chats encrypted on your device and passes prompts through without storing them. See Privacy.

Every message is paid in credits, at the price of the model that answers it.