Chat Interface
Navigate the Chat view — messages, sidebar, voice input, and right panel.
The Chat interface is the primary view for conversational AI interaction. Access it via the sidebar or Cmd + 1 / Ctrl + 1.
Layout
The Chat view has four main areas:
+----------------------------------------------------------+
| Header Bar |
+----------+-------------------------------+---------------+
| | | |
| Chat | Message Stream | Right Panel |
| Sidebar | | (collapsible)|
| | | |
|(sessions)| | |
+----------+-------------------------------+---------------+
| Composer |
+----------------------------------------------------------+Header Bar
The header bar contains three sections:
Left:
- Sidebar toggle (show/hide session list)
- Workspace selector dropdown (switch between project contexts)
- New Chat button (+ icon,
Cmd + N)
Center:
- Mode Selector: Transcribe | Optimize | Chat
- Profile Selector: Current optimization profile (searchable dropdown)
- Response Profile: Fast or Deep
Right:
- Microphone device selector
- Global search (searches sessions, transcripts, prompts, artifacts, agents)
- Settings button
- Right panel toggle
Chat Sidebar (Left Panel)
The left panel lists your chat sessions:
- New Chat button at the top
- Sessions grouped by workspace
- Each session shows: name (editable) and last updated timestamp
- Session actions: Right-click or hover to rename or delete
- Bulk selection: Checkboxes for multi-select, with Select All / Clear / Delete Selected
- Collapse/expand with the toggle button
Message Stream
User Messages (right-aligned)
- Your text content
- Voice input badge (microphone icon if the message was spoken)
- Attachment chips (file icon + filename, for attached files)
- Context blocks (dark cards showing injected context, with edit/remove buttons)
Assistant Messages (left-aligned)
- Colored avatar with the agent's domain icon
- Reasoning panel (collapsible) — shows the AI's thinking process and duration
- Web sources panel — when web search was used, shows source cards with favicons
- Content rendered in rich markdown (code blocks with syntax highlighting, tables, lists)
- Tool progress capsule during streaming — shows what the AI is doing ("Searching...", "Step 1 of 3")
- Metadata row: Agent name, profile used, mode, processing time
- Action buttons: Copy, Regenerate, Save to Library (bookmark icon)
- Aborted state: If you stop generation, shows yellow tint with "(stopped)" label
Streaming Responses
When the AI responds, text appears in real-time (streaming).
To stop generation: Click the Stop button (red square) that appears in the composer area during streaming.
Response Profiles
Control response detail level with two options in the header bar:
| Profile | Icon | Temperature | Use Case |
|---|---|---|---|
| Fast | Green lightning bolt | Lower | Quick answers, concise code |
| Deep | Purple brain | Higher | Detailed analysis, thorough explanations |
Message Actions
Hover over any assistant message to see:
- Copy — Copy the message content to clipboard
- Regenerate — Re-run with the same input
- Save to Library — Bookmark this message for later reference
Voice Input Modes
Vocoding offers two voice input behaviors:
Auto-Send Mode (Default)
Indicated by a green "A" badge on the microphone button.
- Click microphone or press hotkey
- Speak your message
- Recording automatically stops (with VAD) or click stop
- Message is sent immediately after transcription
Best for: Quick voice commands, rapid workflow
Draft Mode
Indicated by an orange "D" badge on the microphone button.
- Click microphone or press hotkey
- Speak your message
- Recording stops
- Transcription appears in the composer for editing
- You review/edit, then press Enter or click Send
Best for: When you want to review before sending, complex instructions
Switching Between Modes
- Shift + click the microphone button to toggle
- Or hover over the microphone to see the mode menu
Microphone Button States
| State | Appearance | Meaning |
|---|---|---|
| Idle | Microphone icon, neutral color | Ready to record |
| Recording | Red background, square stop icon, timer, waveform | Actively listening |
| Processing | Spinner icon, disabled | Transcribing your audio |
| Error | Warning icon with badge | Something went wrong (auto-dismisses after 5s) |
Right Panel
The right panel has 4 tabs, accessible via icons at the top. Toggle the panel with the button in the header bar.
Context Tab
Shows your current project context:
- Workspace info: Project name, root path, alternate app path, git branch
- AI Config files: CLAUDE.md, .cursorrules, etc. — each toggleable, with a global master toggle
- Tech stack chips: Framework, language, package manager (with icons)
- Runtime context: Active application name, attached files count
Artifacts Tab
Displays generated artifacts (code, text, documents) from AI responses:
- Grid of cards, each showing:
- Type badge (colored: code, text, PRD, spec, tasks, image)
- Icon, title, language label, line count
- Preview (first 200 characters)
- Actions per artifact: Copy, Save to Library, Save As (file picker), Write to File
Settings Tab
Quick access to provider and model configuration without leaving the chat.
Inspect Tab
Developer-focused information:
- Agent stack: Selected agent, current mode, lock status
- Token usage: Session tokens, context tokens, total
- Activity logs: Timestamped entries with type badges (info, agent, context, warning)