TUI Full-Screen Mode

niuma_code's default interface is a full-screen terminal UI built on prompt_toolkit, optimized for long coding sessions. The input box stays fixed at the bottom — you can append new messages at any time during LLM streaming output.

Layout

┌──────────────────────────────────────────────┐
│           Output Area (scrollable)            │
│                                              │
│  [Conversation messages + Thinking content]   │
│                                              │
├──────────────────────────────────────────────┤
│ [1/3] "current message"                      │  ← Queue bar
│──────────────────────────────────────────────│
│ > Type your message here...                  │  ← Input box
│──────────────────────────────────────────────│
│ [|] model · high ~1234 out / ~5678 in        │  ← Status bar
└──────────────────────────────────────────────┘

Layout layers (top to bottom):

Output Area — auto-scrolling conversation display with ANSI color rendering
Queue Bar — shows [N/M] "current message" when messages are queued, or ── separator
Input Box — single-line input with autocomplete, expands up to 10 lines with word wrap
Separator — dynamic-width divider
Status Bar — real-time streaming status (spinner, tokens, model, effort)

9 Modal Overlays

Access overlays via keyboard shortcuts or commands:

Shortcut	Overlay	Access Method	Purpose
`/model`	Model Settings	Command only	Switch provider/model/effort via visual popup
`Ctrl+Q`	Message Queue	Keyboard	View and reorder pending messages
`/context`	Context Switch	Command	Multi-context management
`/messages`	Messages	Command	View, edit, merge, star conversation rounds
`/ide`	IDE Drafts	Command	Browse and select Python orchestration drafts
`/tag`	Tag Management	Command	Manage custom #tags
`/lang`	Language	Command	Switch UI language
Auto	Permission Confirm	Auto-trigger	Approve or deny tool execution
`/ide new`	Editor	Command	Exit TUI → external editor → return

Note: Ctrl+M does NOT open Model Settings. In prompt_toolkit, Ctrl+M == Enter (both send 0x0d). The model popup is command-only via /model.

Model Settings (`/model`)

Three-column navigation: Provider → Model → Effort.

Key	Action
`↑/↓`	Move focus between columns
`←/→`	Cycle values within current column
`Enter`	Apply selection
`Esc`	Cancel (restore snapshot)

Message Queue (`Ctrl+Q`)

Key	Action
`↑/↓`	Select message
`Shift+↑/↓`	Reorder messages
`Delete`	Remove selected message
`Esc`	Close panel

Context Switch (`/context`)

Key	Action
`↑/↓`	Navigate contexts
`Enter`	Switch to selected context
`Delete`	Evict (remove) selected context
`Esc`	Close panel

Messages (`/messages`)

Full conversation management with multi-select editing.

Key	Action
`↑/↓`	Navigate units
`Ctrl+↑/Down`	Expand selection
`Alt+↑/Down`	Move unit position
`Delete` or `d`	Delete selected
`m`	Merge selected (LLM summarizes)
`c`	Clear all recall injections
`Enter`	Star/unstar round
`Ctrl+S`	Save changes to disk
`Esc`	Undo changes / Close

IDE Drafts (`/ide`)

Key	Action
`↑/↓`	Navigate drafts
`Enter`	Open selected draft in editor
`Delete` or `d`	Delete draft
`Esc`	Close panel

Permission Confirm (auto)

When a tool needs approval, a centered popup appears with 4 options:

Key	Option
`1` or `↑/↓ + Enter`	Allow (this time)
`2`	Always allow (session)
`3`	Always allow (tool type)
`4`	Deny

Keyboard Shortcuts

Global

Shortcut	Action
`Enter`	Submit message
`Tab`	Trigger/accept autocomplete
`Ctrl+Enter`	Insert newline (without submitting)
`Esc`	Cancel current generation
`Ctrl+C`	Copy selected text (or cancel if nothing selected)
`Ctrl+D`	Exit program
`Ctrl+R`	Toggle memory recall on/off
`Ctrl+P`	Copy last LLM reply to clipboard
`Ctrl+I`	Open IDE mode (`/ide`)
`Ctrl+G`	Open system editor for input
`Ctrl+S`	Save messages (when `/messages` panel is open)
`PageUp/PageDown`	Scroll output area by 10 lines

History Navigation

Key	Action
`↑`	Previous input (when no overlay open)
`Down`	Next input (when no overlay open)

Input Prompt Indicator

The input box shows different prefixes based on state:

> — Normal mode
🧠 — Memory recall enabled

Autocomplete System

Type special prefixes to trigger context-aware completions:

Prefix	Triggers	Example
`@`	File selector (fuzzy search)	`@app/main.py`
`/`	Command completion	`/model`, `/loop`
`#`	Custom tag text expansion	`#plan_mode`, `#skip_mode`, or user-defined tags

Completions appear as you type. Press Tab or Enter to accept, keep typing to filter.

Mouse Support

Action	Behavior
Scroll wheel	Browse conversation history
`Ctrl+Scroll`	Adjust font size (Windows)
Left-click drag	Select text in output area
Left-click release	Auto-copy selected text to clipboard
Right-click	Paste from clipboard (or copy selection)

Mouse support is active in the output area for text selection and scrolling. The output area highlights selected text with a distinct background color.

Status Bar

The status bar renders real-time information during LLM generation:

[|] model · high ~1234 out / ~5678 in

7 Status States

State	Display	When
Idle	Final stats or empty	Between generations
Thinking	`[spinner] 思考中... \| model · effort`	Waiting for first token
Generating	`[spinner] ~N out / ~M in \| model · effort`	Streaming content
ToolExecuting	`[spinner] tool_name (phase)`	Running a tool call
Compressing	`[spinner] [==== ] 40% 压缩中`	Context compression
WaitingForCompression	`[spinner] 等待压缩完成`	Async compression pending
Error	`Error: message` (auto-clears in 5s)	API/network errors

The spinner cycles through [|] [/] [-] [\] at 100ms intervals during active states.

Input Features

Commands

Type / at the input to see available commands:

/model         Open model selection popup
/context       Multi-context switch (new/rename/del)
/clear         Clear conversation history
/compact       Manually compact context
/recall        Toggle memory recall [Ctrl+R]
/loop          Engineered task completion (plan→run→verify)
/copy          Copy reply to clipboard [Ctrl+P]
/ls            List files (fuzzy search)
/cat           Read file content
/edit          Open system editor [Ctrl+G]
/refactor      Refactor code
/ide           Python orchestration mode [Ctrl+I]
/new           Start new topic
/messages      Conversation management
/star          Star/unstar conversation round
/tag           Custom #tag management
/brainstorming Brainstorming design skill
/planning      Planning-with-files skill
/help          Show help
/restart       Restart service
/quit /exit    Exit program [Ctrl+D]

Model Switching

Use the /model command to open a TUI overlay or CLI list for model selection:

/model    → Open model selection overlay (TUI) or list all providers (CLI)

Type # followed by a tag name to trigger text expansion (not model routing):

#plan_mode    → Expand plan_mode tag
#skip_mode    → Expand skip_mode tag

Output Rendering Pipeline

LLM Stream → _raw_buffer → Rich Console → StringIO → ANSI parse → FormattedText → Output Area

ANSI escape codes are parsed into prompt_toolkit FormattedText tuples, preserving Rich colors
Terminal resize triggers dynamic separator width updates and output re-wrapping
Context switching replaces the entire output area content (per-context output isolation)

Accessibility

Font scaling: Ctrl+Mouse Wheel for font size adjustment
Keyboard-only: Full functionality without mouse
High contrast: Works in most terminal color schemes
Wide terminal: Responsive layout up to 200+ columns
CJK support: Proper width calculation for double-width characters