Chat Interface
The chat interface is the main working surface in RealTimeX. It is where you ask questions, attach files, invoke agents, switch models, use voice controls, and review answers with citations or follow-up suggestions.
Not every workspace shows the exact same controls at all times. The current chat surface changes based on:
- whether the thread is empty or already has history
- whether an agent session is active
- whether voice chat is open
- whether the workspace uses a special flow such as meeting capture
- whether some controls are collapsed into a compact
More optionsmenu
Main areas
Most chat threads are organized around three areas:
- the message history
- the prompt composer at the bottom of the screen
- context and action pickers that open above the composer when needed
On a new or empty thread, RealTimeX can also show a landing state with quick links, a centered composer, and suggested agent shortcuts before the first message is sent.
Message history
The message history is where RealTimeX renders user prompts, assistant replies, attachments, file previews, citations, and follow-up suggestions.
Depending on the content, the current interface can show:
- regular text messages
- attachments and inline file previews
- image or audio blocks
- assistant citations when a reply includes sources
- suggested follow-up questions after the latest assistant answer
Some assistant messages also include extra context from the current workspace, retrieval pipeline, or active agent session.
Message actions
Actions vary by message type.
All standard text messages support:
- copy
Assistant replies can additionally show:
- speak message
- regenerate the latest assistant reply
- thumbs-up feedback
- citations when sources are available
- add to knowledge
- a
Moremenu for actions such as forking the thread or deleting the message
These controls appear only when they make sense. For example, Regenerate is tied to the latest assistant response, and Citations appears only when RealTimeX has source data to show.
Prompt composer
The prompt composer is the control bar at the bottom of the chat screen. This is where you prepare the next turn.
The current composer can include:
Attach fileWeb SearchAgentsTerminalKnowledgesCommandsLLM SelectorSpeech-to-TextVoice ChatSendorStop
Depending on screen size and layout, some of these controls may move into a compact More options menu instead of staying visible as dedicated buttons.
Files and context
Use the paperclip button to add files to the current conversation.
In the current product, files can be used in more than one way:
- added as prompt context for the current chat only
- uploaded and embedded into the workspace for later retrieval
- attached as images or files directly to the current turn
The attachment row above the composer shows each item's state, such as uploading, chat-only context, or embedded workspace content. You can also remove queued items before sending the message.
Commands and agents
The chat interface supports both regular chat commands and agent-driven sessions.
Commands
Use the Commands button or type / to open the slash-command surface.
In normal chat, the current command surface can include:
- built-in commands such as
/reset - image-generation entry points such as
/image - saved slash-command presets
If an agent session is active, the command surface changes. The current UI can show:
/exitto end the session- extra commands advertised by the active agent
For the full preset management and usage guide, see Slash Commands.
Agents
Use the Agents button or type @agent in the prompt to work with agents.
Some workspaces or threads can also preselect a default agent for you.
When an agent session is active:
- the prompt area can wait until the agent explicitly requests input
- the available command set changes
- terminal-linked or browser-linked workflows can surface extra controls
For deeper agent behavior, see Using AI Agents and Agent Runtime.
Terminal-linked chat
When a thread is linked to a terminal-backed agent session, the composer can show a terminal link badge above the editor.
Use it when you need to:
- see which terminal session the thread is currently linked to
- relink the thread to a recent session
- detach the chat from the current terminal session
This matters most for CLI-backed agents and runtime workflows.
Goal tracking actions
The current chat surface can also hand durable work off to the Ambient Agent board.
In regular workspaces, a thread can show:
Track this threadTrack as goal
Use this when a conversation has become real work that should be followed across time instead of staying trapped inside one thread.
Terminal session history cards can also show:
Track in Ambient
Use this when a coding run, terminal task, or CLI-backed workflow should become an ongoing goal with later review, monitoring, or intervention.
Both actions create or attach a goal and then route you into the Ambient Agent workspace so you can manage it from the central board.
For the full goal workflow, see Goals & Ambient Dashboard.
Speech and voice
The current chat surface supports two different speech flows.
Speech-to-Text
Use the microphone button to dictate a prompt.
RealTimeX currently supports multiple speech-to-text backends, and the chat input can respect language preferences as well as microphone permissions on web or desktop.
Voice Chat
When the prompt is empty, the send button area can switch to a Voice Chat launcher instead.
Use this mode when you want a hands-free conversation flow instead of typing one message at a time.
For output audio, the message-level Speak action uses your configured text-to-speech setup rather than the voice-chat session itself.
See Voice Chat, Text-to-Speech, and Transcription Models.
Chat preferences
Open Settings > Chat for chat-wide behavior preferences.
Some chat behavior is configurable in settings, including options such as:
- auto-submit
- auto-speak
- spell check
- scrollbar visibility
- agent notification behavior
These preferences affect how the chat surface behaves, not just how it looks. Workspace-specific chat rules such as model selection, prompts, and retrieval settings are configured separately inside each workspace.