Recall

› What it does

• Spans a phone, a laptop and an Arduino Uno. It captures audio, text, browser content and images into one persistent memory.
• On-device inference through ONNX Runtime, accelerated by the Snapdragon NPU.
• Hybrid RAG keeps retrieval fast as context grows. If nothing matches, it says so instead of guessing.
• A single RECALL_BACKEND flag runs the same code on the NPU, on any laptop through Ollama, or fully offline.
• The hardest part: Python on Windows ARM64, where httptools, cryptography and sqlite-vec shipped no wheels.