DigUp

Search your computer by what's in your files. Describe a photo, a clause in a lease, a minute of a podcast or a moment in a video, and DigUp opens the file at that page or that second. Google DeepMind's EmbeddingGemma 2 runs on your own machine through llama.cpp, so nothing you index leaves it. On the Mac, with Windows and Linux in beta.

mlx-tune

Bringing the Unsloth experience to Mac users via Apple's MLX framework. Fine-tune LLMs and vision-language models locally on Apple Silicon with LoRA / QLoRA support.

mlx-dspark

DeepSeek's DSpark and z-lab's DFlash speculative decoding — running natively on Apple Silicon via MLX.

kaggle-tpu-lab

Run open models on Kaggle's free TPU v5e-8 behind a public endpoint that speaks the OpenAI and Anthropic APIs, ready for Claude Code, Codex CLI or opencode. Qwen3.8-27B in full bf16 at ~130 tok/s, and GLM-5.3-Flash (320B MoE) at ~64 tok/s on a JAX engine I wrote for it.

Lekho — লেখো

Avro Phonetic Bangla keyboard for macOS — native Apple Silicon, works on M1 through M5. Free, open-source replacement for iAvro.

omniASR-server

OpenAI-compatible ASR/STT API server powered by Meta's omnilingual-asr model. Real-time streaming over WebSocket, batch transcription over REST.

supertonic-server

Local OpenAI-compatible TTS server for Supertonic-3 — sentence-level streaming across 31 languages. Pipecat/LiveKit-ready, runs on CPU, CoreML, or CUDA.