You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
EmotiBit support: connect an EmotiBit board over Bluetooth (emotibit-ble firmware) or WiFi (stock firmware). Heart rate, inter-beat intervals, SpO2 and skin temperature record as device vitals, accelerometer and gyroscope as motion, and battery as battery. A BLE board is found by its GATT service; add a WiFi board as emotibit:<id or ip>.
Electrodermal activity and other scalar sensor streams are recorded: a new _sensors sidecar (CSV or Parquet) stores one timestamp_s, channel, unit, value row per sample, so EDA, the SCR metrics, thermopile, humidity, magnetometer and raw PPG share one file regardless of rate. Subscribers also receive a live SensorSample event. Other devices with a scalar stream and no dedicated event can use the same DeviceEvent::Sensor.
EDA analysis: skin conductance is analysed as it is recorded. Every five seconds you get the tonic level (SCL), skin conductance responses per minute and their size, and arousal as a z-score against the session's own baseline. Artifacts (lost contact, impossible jumps) are rejected rather than counted. The results are stored beside the raw samples as eda_* channels.
Live EDA card: the dashboard shows the skin-conductance trace with the derived metrics, for an EmotiBit or any device that reports an eda channel. It says when the baseline is still being learned and warns about poor skin contact.
Bugfixes
Sidecar files were miscounted and left behind: _vitals and _fnirs files were treated as sessions by history listing, loaded as EEG by the EEG-file filter, and not removed when a session was deleted. One shared list of sidecar suffixes now drives all three.
Build
npm run release -- --rc --skip-checks (and npm run bump -- --skip-checks): cut a release candidate without any local testing or compilation. bump skips its preflight (frontend checks, the i18n check, vitest, clippy and cargo test), and npm run tag pushes the tag with --no-verify so the pre-push hook does not re-run them either. The version and tag safety checks and the pre-commit hook still run, and CI is the first to see the commit.
Heavy integration tests are opt-in: nine model-, GPU- and measurement-dependent test targets (fft_*, qwen3_tts_e2e, tts_validate_e2e, asr_e2e, asr_funasr_nemotron_smoke, rlx_image_embed, rlx_semantic) sit behind a per-crate slow-tests feature, so cargo test --workspace no longer links them. Most were #[ignore]d or skip themselves without cached models. cargo test-slow runs everything; the documented run commands and the two TTS scripts pass the feature.
Smaller dev builds: dependency crates build with no debug info and DuckDB at opt-level = 1, split-debuginfo = "off" stops kache writing .dSYM bundles, the skill library is built as an rlib only, benchmark binaries sit behind a bench-tools feature, and seven crates link their fast integration tests into one it binary each. A full workspace test build is about 6 GB instead of over 49 GB.
Linux release and CI builds no longer fail on a stale apt index: the runner image's package index can be days old and the mirror drops superseded versions, so cache-apt-pkgs-action asked for files that were gone, the whole install failed while the step still reported success, and the build died later on glib-2.0 was not found (v0.1.1, v0.1.1-rc.3 and v0.1.2-rc.1) or libsoup-3.0 not found (PR CI). The apt index is now refreshed first, with retries, and a check right after the install fails immediately with a clear message if glib, GTK, libsoup or WebKit is missing.