60-Second Quickstart

System Prerequisites

Howitzer is compiled natively with Qt 6.5.3 for zero-overhead performance:

• macOS: macOS 13 (Ventura), 14 (Sonoma), or 15 (Sequoia). Apple Silicon (M1-M4) or Intel x86_64.
• Linux: Ubuntu 22.04+, Debian 12+, Fedora 38+, Arch Linux. (GLIBC 2.31+).
• Windows: Windows 10/11 64-bit with DirectX 12 support.
• RAM: 8 GB unified memory for cloud APIs; 16 GB+ recommended for running 7B-14B local GGUF models.

Install & Launch

Install via direct binary download or one-line terminal command:

On macOS, drag Howitzer.app to /Applications and launch. There is no mandatory account registration or onboarding wizard—the tactical workbench opens directly into DirectFire.

Connect Frontier Cloud Models (Optional)

If you use frontier models (OpenAI, Anthropic, Gemini, Mistral), navigate to Magazine [MZ] in the top toolbar:

• Click [ + ADD SECRET ] and select your provider.
• Paste your raw API key. It is immediately encrypted using AES-256 via macOS Keychain or Linux libsecret.
• Your keys are never transmitted to any third-party server or telemetry collector.

Download Your First 1-Click Local Model

To run 100% private, offline inference with zero API costs:

• Switch to the Firing Range [FR] or Arsenal [AR] tab.
• In the model dropdown, select Local Model Hub.
• Click [ RUN 1-CLICK ] next to Qwen 2.5 Coder 7B or DeepSeek R1 Distill.
• Howitzer downloads the pre-quantized Q4_K_M GGUF directly to your local application support cache and initializes Metal GPU offloading.

Execute Your First Tactical Salvo

Switch to Firing Range [FR]:

• Enter a prompt in the central salvo input (e.g., "Write a C++ thread-safe ring buffer").
• Assign Model A to Gemini 3.5 Flash, Model B to Local Qwen 2.5 Coder, and Model C to Claude 3.5 Sonnet.
• Click the [ PULL SALVO ] tactical action button.
• Observe side-by-side token streaming, microsecond latency telemetry, and unified token diffing in real time.