Chat

Chat is the core end-user workflow in OrcAI. You can start chats from scratch, start from a bot, attach sources, switch branches, and tune per-chat settings.

Entry Points

• Sidebar action: Start a chat
• Chat index page: Start a chat
• Direct path: /app/chat/setup

The setup page offers two options:

• Use a bot: open a selector dialog and choose from available bots.
• Start fresh: no preselected bot.

Both paths go to the new chat composer where you send the first prompt.

Starting A New Chat

Choose A Starting Point [step]

Use a published bot when you want a predefined behaviour and knowledge setup. Start fresh when you want to choose the model and sources directly.

Configure The First Message [step]

Before sending the first message, you can configure:

• Bot (optional)
• Model and provider (required)

Send And Continue [step]

When you send the first message, OrcAI creates the chat and opens the conversation view.

Chat List

The chat list supports:

• Search by text
• Filter by bot
• Sorting and pagination
• Table actions (bulk actions depend on capabilities)

In-Conversation Features

In an active chat, OrcAI supports:

• Streaming model responses
• Message actions (for example regenerate and edit, depending on message type)
• Usage metadata on assistant messages
• Download conversation
• Chat tour trigger

Branching

If a chat has multiple branches, a branch switcher appears at the top-right of the conversation.

• Switching branch updates both active branch state and URL query.
• Branch names are reflected in the page title when not the default Main branch.

Attachments In Chat

Chat supports two attachment sources:

• Local uploads during composition
• Existing content from the content library

Current hard limits and accepted formats in the app settings:

• Max attachments per message: 8
• Accepted patterns: image/, application/pdf, text/, DOCX, PPTX

The send button allows either:

• Text only
• Attachments only
• Text plus attachments

Chat Settings

Each chat has a settings dialog where you can configure:

• Model/provider
• System prompt (only for chats not locked to a bot)
• Advanced generation settings
  • temperature
  • top-p
  • max tokens
  • frequency penalty
  • presence penalty

For bot-linked chats:

• System prompt is locked to the bot template block
• Attached blocks are managed by the bot, not directly in chat settings

For non-bot chats:

• You can attach one behaviour block and multiple content collection blocks directly to the chat

Quota Badge

When quota data is available, the chat header shows remaining amount for the assigned pool.

The badge can represent either:

• Requests
• Tokens

Retrieval And Citations

Responses can include tool-driven retrieval output and source-aware rendering (for example citations and knowledge lookups) through message part renderers.