⚡ LLM & Image AI TUI: Provider Manager & Chat/Image Laboratory

English | 简体中文

llm_tui is a fast, modern, terminal-based user interface (TUI) application written in Go for auto-probing, managing, and interactively testing LLM & Image AI Provider API endpoints (supporting OpenAI Chat API, OpenAI Responses API, Anthropic Messages API, and OpenAI Images API).

Built using the Charm.sh ecosystem (bubbletea, lipgloss, viewport, textarea) and a pure-Go SQLite database (modernc.org/sqlite, CGO-free), llm_tui allows developers and AI engineers to quickly verify provider connectivity, benchmark latency/tokens, generate images, and debug custom JSON request payloads side-by-side.

---

✨ Features

• 🔍 Automated Capability Probing & Setup Wizard:
  • Connects to any OpenAI or Anthropic compatible Base URL.
  • API keys are optional: leave empty for unauthenticated providers / local servers.
  • Automatically queries the /models endpoint (with root-URL fallback) to discover available models; if discovery fails, prompts you to enter the exact model ID manually.
  • Smart Image Model Direct Routing: When selecting image generation models (e.g. image, dall-e, flux, imagen), the probe network check is automatically skipped to preserve your compute quota and prevent 429 rate limit errors, taking you straight to the Image Laboratory.
  • Concurrently probes text endpoints with realistic timeouts to detect active capabilities (openai_chat, openai_responses, anthropic_messages).
• 🖼️ Dedicated AI Image Laboratory:
  • Purpose-built interactive laboratory for openai_images models.
  • Generates with token-saving, lowest-compute preset ("aspect_ratio": "1:1") by default.
  • Alt+A Aspect Ratio Switcher: Interactive selector sorted from smallest to largest compute size (1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9) with instant JSON update and persistence.
  • Auto-Saving & Anti-Lag Optimization: Automatically decodes Base64 data and writes image files to ./generated_images/ (YYYYMMDD_HHMMSS_<prompt_slug>.png/.jpg) while rendering concise summaries in the viewport to keep the terminal smooth.
  • Ctrl+O Quick Open: Press Ctrl+O to instantly open the latest generated image in your operating system's default viewer.
• 💾 Pure Go SQLite Storage (CGO-Free):
  • Local database (providers.db) located alongside the binary.
  • Stores provider configs, custom payload templates, model choices, and reasoning preferences.
  • Auto-fills saved API Keys when matching Base URLs.
• 🧪 Side-by-Side Split-Pane Chat & Image Laboratory:
  • Left Pane (Request Payload Editor): Live multi-line JSON payload editor with active focus. Press Ctrl+S to send requests instantly without leaving the editor.
  • Right Pane (Response Viewport): Displays HTTP status code, latency, token usage (Prompt, Completion, Total), tokens-per-second (TPS) throughput, and formatted JSON response with smooth PgUp / PgDn page scrolling.
• 🤖 Embedded Model Switcher:
  • Press Alt+M in the Chat Laboratory to open an embedded model picker viewport in the left panel.
  • Displays up to 20 models with centered scrolling pointers (👉). Selection updates model choice and saves to SQLite automatically.
• 🧠 Flexible Reasoning Effort Shortcuts:
  • Press Alt+1 to Alt+4 to toggle Reasoning Effort (none, low, high, max).
  • Automatically formats reasoning_effort for OpenAI Chat, reasoning for OpenAI Responses, and output_config.effort for Anthropic Messages.
• ⚡ Real-Time SSE Streaming & Reasoning Process Visualization:
  • Native support for typewriter-style real-time streaming via HTTP SSE (text/event-stream).
  • Auto-detects "stream": true/false payload mode (toggle it manually anytime with Alt+S): Pretty JSON for non-stream, live typewriter display for stream mode with dedicated 💭 Thinking Process and 💬 Response Content cards.
• 🎨 Terminal Glamour Markdown Rendering & Code Syntax Highlighting:
  • Integrated Charm glamour Markdown renderer engine.
  • Automatically highlights code blocks (Python, Go, SQL, JSON, etc.), header tags, bold text, and lists.
• 📐 Responsive Word Wrap & Auto Reflow:
  • Dynamically calculates Viewport width and performs real-time text reflow during terminal window resizing, eliminating horizontal border overflow.
• 🔑 Cross-Platform Lossless Unicode & UTF-8 Clipboard (OSC 52 + Native):
  • Supports both local desktops (X11 UTF8_STRING / Wayland wl-copy, macOS pbcopy, Windows PowerShell) and remote SSH / headless Linux environments (OSC 52 ANSI escape sequences).
  • Copies text directly into your local machine's clipboard without needing xclip or X11 forwarding over SSH.
  • Solves Chinese and Emoji clipboard garbled text issues natively.
  • Ctrl+Y: One-click copy Request Payload; Ctrl+U: One-click copy Response JSON.

---

🚀 Installation & Quick Start

Option 1: Download Pre-built Release Binaries

Download the archive for your OS/Architecture from the GitHub Releases (each archive contains the llm_tui binary, or llm_tui.exe on Windows):

• Linux (x86_64): llm_tui-<version>-linux-amd64.tar.gz
• Linux (ARM64): llm_tui-<version>-linux-arm64.tar.gz
• macOS (Apple Silicon): llm_tui-<version>-darwin-arm64.tar.gz
• macOS (Intel): llm_tui-<version>-darwin-amd64.tar.gz
• Windows (x86_64): llm_tui-<version>-windows-amd64.zip

Extract and run:

./llm_tui

Option 2: Build from Source

Requires Go 1.25+:

git clone https://github.com/GreyRaphael/llm_tui.git
cd llm_tui
go build -o llm_tui .
./llm_tui

---

⌨️ Keybindings Reference

Provider Manager View

Key	Action
n	Add New Provider (Launch Setup Wizard)
Enter / t	Open Selected Provider in Chat/Image Laboratory
d	Delete Selected Provider Record (press twice to confirm)
↑ / k, ↓ / j	Navigate Provider Cards
q	Quit Application

Setup Wizard View

Key	Action
Tab / Shift+Tab	Switch Input Fields
↑ / k, ↓ / j	Select Model or API Type
Enter	Proceed to Next Step / Confirm Selection
Esc	Cancel and Return to Provider Manager

Chat & Image Laboratory View (Split-Pane)

Key	Action
Ctrl+S	Send Request Payload & View Response / Generate Image
Ctrl+O	Open Latest Generated Image in System Default Viewer
Alt+A	Toggle/Select Image Size & Aspect Ratio Presets (Supporting 0.5K/1K/2K/4K dimensions sorted smallest to largest)
Alt+S	Toggle Streaming Mode (stream: true/false, chat models only)
Tab / Shift+Tab	Switch Focus between Left Pane (Request) & Right Pane (Response)
Alt+M	Toggle Embedded Model Switcher Menu
Alt+1 - Alt+4	Switch Reasoning Effort (1: none, 2: low, 3: high, 4: max)
Ctrl+Y	Copy Request JSON Payload to Clipboard
Ctrl+U	Copy Response JSON Payload to Clipboard
PgUp / PgDn	Scroll Response Viewport Page by Page
Esc	Return to Provider Manager

---

🧱 Architecture & Tech Stack

• UI Framework: github.com/charmbracelet/bubbletea
• Styling: github.com/charmbracelet/lipgloss
• Components: bubbles/textarea, bubbles/viewport, bubbles/spinner
• Database: modernc.org/sqlite (Pure Go SQLite driver, NO CGO required)
• Clipboard & Terminal Protocol: github.com/aymanbagabas/go-osc52 (OSC 52 escape sequences for remote SSH clipboard) & native cross-platform clipboards

---

📄 License

MIT License. Developed for pairing with AI coding agents and LLM endpoint benchmarking.