Fullmoon

Fullmoon is an on-device chat application for Apple platforms that runs quantized language models directly on local hardware using Apple's MLX Swift framework. By executing inference locally with Metal acceleration, prompts and chat logs remain on the physical device without reaching external networks. It gives developers a private sandbox for testing compact models like Llama 3.2 and DeepSeek-R1-Distill without subscription tiers or API tokens.

Why Fullmoon instead of ChatGPT

ChatGPT routes every prompt, code snippet, and configuration file through OpenAI cloud infrastructure. For engineers handling proprietary codebases, internal architecture configs, or customer data under confidentiality agreements, piping queries to hosted endpoints introduces compliance and leak risks. Fullmoon confines the entire inference loop to the device. Computations run locally within unified memory on Apple silicon, ensuring prompts never traverse third-party proxies and work remains uninterrupted on flights or in air-gapped network segments.

The operational trade-off lies in model scale. ChatGPT connects to frontier reasoning engines with hundreds of billions of parameters, broad web browsing, and multi-file code synthesis. Fullmoon runs edge-quantized weights between one and three billion parameters. These smaller models handle targeted syntax conversions, bash regex drafting, and text summarization, but cannot solve intricate architectural refactors. In return, Fullmoon integrates into local workflows via Apple Shortcuts, enabling script automation without managing token billing, API quotas, or network latency.

What it does well

• On-device MLX acceleration: Runs 4-bit and 8-bit quantized weights via Apple's MLX Swift framework and Metal 3, achieving responsive token generation without network latency.
• Apple Shortcuts integration: Exposes native shortcuts actions that let engineers pipe local model outputs directly into shell scripts, clipboard actions, or system automation.
• Air-gapped offline operation: Processes queries without internet connectivity, allowing DevOps engineers to draft scripts and verify syntax during travel or inside restricted environments.
• Custom system prompts: Supports tailored system instructions to enforce output structures like pure JSON, regex patterns, or shell command snippets.
• Zero credential overhead: Eliminates API key rotations, token meters, and monthly seat subscriptions by running self-contained weights stored on disk.