Verbodus is a fast and lightweight desktop application designed to benchmark the performance and generation latency of Large Language Models (LLMs) in real time. It connects to any OpenAI-compatible API endpoint (such as local engines like Ollama, LM Studio, and vLLM, or remote cloud services) to evaluate throughput, prefill latency, and decode speed with high precision.

Built with Tauri v2, Vue 3, and Chart.js, Verbodus features a modern glassmorphism user interface designed for developers who want to inspect and compare model execution speeds locally on their machines.

• Performance Playground: Execute custom prompts and watch streaming responses generate in real time with an interactive caret and live telemetry stats.
• KPI Metrics Dashboard: Tracks standard industry benchmarks with color-coded performance indicators:
  • TTFT (Time to First Token / Prefill Latency): Measures prompt processing responsiveness.
  • TPOT (Time Per Output Token / Decode Latency): Measures average token-by-token generation latency.
  • TPS (Tokens Per Second / Throughput): Measures overall streaming generation speed.
  • Token Auditing: Displays counts for Prompt, Completion, and Total tokens.
• Real-Time Throughput Curve: Renders a dynamic line chart of generation throughput (TPS over time) as token chunks stream in.
• Engine Comparison Dashboard:
  • Compare up to 4 historical benchmark runs side-by-side.
  • Dual-axis bar chart contrasting TTFT (ms) against Average TPS.
  • Detailed metadata inspector displaying prompt parameters and exact token breakdowns.
• Custom API Profiles: Setup and save configuration presets for local or remote servers. Adjust generation properties like Temperature, Max Tokens, System Prompts, and toggle between Streaming and Non-Streaming modes.
• Persistent Storage: All runs history and API profiles are persisted locally in browser-native storage.

Verbodus automatically color-codes your benchmark results based on latency and throughput thresholds:

• Frontend: Vue 3 (Composition API / <script setup>)
• Styling: Vanilla CSS (Custom properties, CSS variables, glassmorphic filters)
• Charts: Chart.js (Dual-axis charts & real-time line charts)
• Desktop Runtime: Tauri v2 (utilizing native macOS WebKit / WebViews)
• Build System: Vite

To run Verbodus locally, ensure you have:

1. Bun (v1.0 or higher recommended)
2. Rust & Cargo (required for compiling the Tauri desktop client)
3. Local LLM Engine (any OpenAI-compatible API running locally):
   • Ollama: Download & Install (Default API port: 11434)
   • LM Studio: Download & Install (Default API port: 1234)
   • vLLM: Documentation (Default API port: 8000)

1. Clone the repository and navigate to the project directory:

   cd "Verbodus"

2. Install dependencies:

   bun install

To run a fast-loading web-only preview in your default browser:

bun dev

Open http://localhost:1420 in your browser.

To compile the Tauri Rust backend and launch the native desktop window:

bun tauri dev

To package the final, production-ready desktop bundle:

bun tauri build

This compiles the Rust backend in release mode and bundles the Vue assets. The final .app or installer will be located in src-tauri/target/release/bundle/.

Verbodus comes pre-configured with popular local LLM defaults:

• Ollama:
  • API URL: http://localhost:11434/v1
  • Default Model: llama3
• LM Studio:
  • API URL: http://localhost:1234/v1
  • Default Model: lmstudio-community
• vLLM / Local Engine:
  • API URL: http://localhost:8000/v1
  • Default Model: meta-llama/Meta-Llama-3-8B-Instruct

Note: For engines requiring authentication (such as remote OpenAI, Anthropic-compatible proxies, or Groq), you can enter your API Key directly in the profile parameters panel.

This project is licensed under the GNU General Public License v3 - see the LICENSE file for details.