TUI & Dashboard
Layout
The terminal UI has three main regions:
- Conversation view (center) — user prompts, assistant responses, tool cards, system notifications, lifecycle events
- Editor (bottom) — wrapped multiline input with history and slash-command palette
- Status footer (bottom bar) — ambient runtime telemetry. Expands into a unified console when deeper instrumentation is active.
Segments
The conversation is rendered as a sequence of typed segments, each with metadata:
| Segment Type | Rendering |
|---|---|
| UserPrompt | Your input, right-aligned with > gutter |
| AssistantText | Response with teal │ gutter, meta header, and markdown highlighting |
| ToolCard | Collapsible card: tool name, args summary, and result preview |
| SystemNotification | System messages such as compaction and errors |
| LifecycleEvent | Single-line status changes and turn transitions |
| TurnSeparator | Thin horizontal rule between turns |
Segment metadata
Assistant responses carry a dim metadata header stamped at creation time, for example:
gpt-5.4 · openai-codex · victory · think:medium Fields include model ID, provider, capability tier, thinking level, and other runtime details.
Tool cards
Each tool call renders as a compact card with the tool name, one-line args summary, and a collapsible result preview.
- Collapsed — tool name, args summary, and a short result preview
- Expanded — larger result body with formatting where appropriate
- Error cards — explicit error styling instead of pretending failure is normal output
Expansion
| Key | Behavior |
|---|---|
Ctrl+O | Pin or unpin the nearest tool card |
Tab | Toggle expand on the nearest tool card when the editor is empty |
Esc | Unpin any pinned segment or close the active popup |
Status footer
The footer is an operator instrument panel, not decorative chrome. It is meant to answer “what engine am I actually on, what state is it in, and how much runway do I have?”
- Context — context class, fill percentage, and token-aware usage display
- Engine — provider identity, selected model, tier, thinking level, and optional limit telemetry
- Memory — durable fact counts and working-memory activity
- System — runtime health, versions, and related state
In the engine panel specifically:
- provider = concrete runtime backend identity, such as
OpenAI/CodexorAnthropic/Claude - model = selected runtime model
- limit = upstream quota or bucket telemetry, when the provider exposes it
That separation is intentional. A Codex bucket name in the limit row is not a second model selector.
Sidebar navigation
Press Ctrl+D to enter sidebar navigation mode and steer the design tree directly.
| Key | Action |
|---|---|
Ctrl+D | Toggle sidebar mode |
Up / k | Navigate up |
Down / j | Navigate down |
Left / h | Collapse node |
Right / l | Expand node |
Enter | Focus the selected node |
Esc | Exit sidebar mode |
Mouse
Mouse capture is on by default. To use your terminal's native text selection,
toggle mouse handling with Ctrl+M. Press Ctrl+Y to copy the selected
conversation segment as plain text.
Editor
The input area is a real wrapped multiline editor with visible cursor placement, prompt history, and proportional height growth.
- Enter sends the prompt
- Shift+Enter inserts a newline
- Arrow keys navigate text and history
Browser surface
/auspex open Launches Omegon's primary browser surface. If the local compatibility bridge is not running yet, Omegon starts it first and then hands the browser launch off to Auspex.
/dash remains available as the local compatibility/debug command for the in-process browser surface.
It is useful for demos, wider live inspection, and compatibility testing, but it is not the primary operator-facing launch path.