Constellation · 02

Mind

FFAI, Mercury, Carmenta, Diana: AI that stays local

Why this band exists

Cloud inference means your voice, your documents, and your camera feed leave the room before the model ever answers. Remade Mind remakes Whisper-class, YOLO-class, and OCR stacks in Rust so the neighborhood can see and hear without renting a GPU in someone else's region.

Why now

"On-device" that still uploads is a lifestyle brand. Households and OEMs need ASR, OCR, and detection that compile into a tab, a phone, or a board: without a Python glue stack and without a training-set commute. Local AI is how Digital Freedom survives contact with a neural net.

How it fits

Media decodes the bytes Mind reads. Memory keeps transcripts and embeddings beside the user. Identity decides who may run the model. Edge puts FFAI next to the sensor. Trust keeps the lockfile under the weights honest. Mind is perception that refuses exfiltration as architecture.

Stars in Mind

Each crate is open under Remade-With-Rust. Open the repo, then the field note, then compose with the rest of the sky.

  • FFAI

    The Remade-with-Rust AI media suite. Engines for ASR, TTS, OCR, VLM, and YOLO-class detection share one Rust surface so perception stays local.

  • Mercury · replaces Whisper.cpp

    FFAI ASR engine: pure-Rust Whisper speech recognition

  • Carmenta · replaces Tesseract / Python OCR

    FFAI OCR engine: text from images without the C tax

  • Diana · replaces YOLO C++ stacks

    FFAI detection engine: object detection in Rust

  • Argus · replaces cloud VLM caption APIs

    Vision-language that captions locally

What to build first

  1. An on-device media suite path through FFAI (ASR, TTS, OCR, VLM, detection)
  2. Speech notes that transcribe in-process with Mercury
  3. Scan and screenshot OCR with Carmenta, including WASM in the browser

Field notes in Mind