mendi

Async Rust library and CLI for streaming fNIRS neurofeedback data from Mendi headbands over Bluetooth Low Energy.

Hardware

The Mendi headband is a consumer fNIRS (functional near-infrared spectroscopy) device that measures blood oxygenation in the prefrontal cortex using infrared and red LEDs. It communicates over BLE using protobuf-encoded messages on six GATT characteristics:

Each optical channel (left, right, pulse) provides three readings:

• IR (infrared) — penetrates tissue to measure oxygenated hemoglobin
• Red — measures deoxygenated hemoglobin
• Ambient — background light for subtraction

Quick start

use mendi::prelude::*;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let client = MendiClient::new(MendiClientConfig::default());
    let (mut rx, handle) = client.connect().await?;

    while let Some(event) = rx.recv().await {
        match event {
            MendiEvent::Frame(f) => {
                println!("temp={:.1}°C  IR: L={} R={} P={}",
                    f.temperature, f.ir_left, f.ir_right, f.ir_pulse);
            }
            MendiEvent::Battery(b) => {
                println!("Battery: {:.2}V charging={}", b.voltage(), b.charging);
            }
            MendiEvent::Disconnected => break,
            _ => {}
        }
    }
    Ok(())
}

CLI usage

# Run the CLI (scans, connects, and prints all data)
cargo run

# With debug logging
RUST_LOG=mendi=debug cargo run

Interactive commands once connected:

• q – quit
• c – write default calibration (auto-cal on)
• e – enable optical sensor
• d – disable optical sensor

TUI (real-time charts)

# Real device
cargo run --features tui --bin mendi-tui

# Simulated (no hardware needed)
cargo run --features tui --bin mendi-tui -- --simulate

Keys: 1 IR view, 2 full view, +/- scale, a auto-scale, v smooth toggle, p pause, r resume, c clear, q quit

Using as a library

[dependencies]
mendi = "0.1.0"

# With simulation/mock support for testing:
mendi = { version = "0.1.0", features = ["simulate"] }

Protocol details

The wire protocol is defined in proto/device_v4.proto (protobuf v3). All characteristics use protobuf encoding. The Frame characteristic (0xABB1) is the primary data stream, delivering real-time sensor readings at the headband's native sample rate.

Frame fields

Convenience methods on FrameReading convert raw IMU values to physical units:

• accel_x_g(), accel_y_g(), accel_z_g() — acceleration in g
• gyro_x_dps(), gyro_y_dps(), gyro_z_dps() — angular velocity in °/s

UUID namespace

Mendi uses a vendor-specific UUID base: fc3eXXXX-c6c4-49e6-922a-6e551c455af5. Use mendi_uuid(short) to construct UUIDs from 16-bit short codes.

Simulation & testing (feature simulate)

The simulate feature adds two test utilities that don't require hardware:

SimulatedDevice

Generates realistic fNIRS data with sinusoidal optical signals, IMU noise, and periodic battery/calibration events:

use mendi::simulate::{SimulatedDevice, SimConfig};
use mendi::types::MendiEvent;

let config = SimConfig {
    frame_rate_hz: 25.0,
    disconnect_after_frames: Some(100),
    ..Default::default()
};
let (mut rx, handle) = SimulatedDevice::start(config);
while let Some(event) = rx.recv().await {
    // Same MendiEvent stream as real hardware
}

Run the simulator binary:

cargo run --features simulate --bin mendi-sim
cargo run --features simulate --bin mendi-sim -- --frames 500 --hz 50

MockDevice

Deterministic scripted mock for unit tests — no timing, no randomness:

use mendi::simulate::MockDevice;
use mendi::types::MendiEvent;

let (mut rx, mock) = MockDevice::with_frames(10, 32);
// Receives: Connected, 10 Frames, Disconnected

MendiHandle API

Once connected, the MendiHandle provides:

License

MIT