psycheros-v0.8.18

Fixed

• Voice WebSocket: 25s ping heartbeat prevents Deno from killing the connection
  during long thinking periods (tool calls, LLM round-trips).
• Vanilla (continuous) voice mode: pttMode is now compared as "ptt" instead
  of truthy, so non-PTT mode no longer blocks audio frame sending at call start.
• Tauri 2.x desktop: native mic capture delivers frames as ArrayBuffer, not
  Array. Both shapes are now handled in the voice WebSocket send path and the
  mic test diagnostic.

---

How to install

The canonical artifact for the harness is the container image:

docker pull ghcr.io/psycherosai/psycheros:<version>
# or `:latest` for the most recent release

See the README's Docker block
for the recommended docker run invocation, env-var reference, and volume
layout.

The archives prefixed Psycheros- below are GitHub's auto-attached
full-monorepo source — useful for building the image yourself, but the published
image at GHCR is the supported install path.

psycheros-v0.8.17

Fixed

• PTT mode: gate audio frame sending and server-side audio on pttHolding to
  prevent spurious RECORDING state
• PTT mode: remove global pttEnabled, make PTT per-call only
• PTT mode: add session.pttMode to track PTT vs vanilla mode
• UI state: always show 'Listening' for idle state regardless of PTT mode
• UI state: show 'Listening' not 'Recording' in vanilla mode
• STT provider: fix module-level sttProvider in fallback path
• VAD: reset nativePeakRms in browser VAD
• VAD: send user_speech_start when VAD detects speech (server STT)

---

How to install

The canonical artifact for the harness is the container image:

docker pull ghcr.io/psycherosai/psycheros:<version>
# or `:latest` for the most recent release

See the README's Docker block
for the recommended docker run invocation, env-var reference, and volume
layout.

The archives prefixed Psycheros- below are GitHub's auto-attached
full-monorepo source — useful for building the image yourself, but the published
image at GHCR is the supported install path.

psycheros-v0.8.16

Fixed

• Silence detector check loop no longer dies when analyserNode isn't ready
  yet. An early return cancelled rescheduling, leaving VAD dead for the entire
  call — same bug class as the WS-connect race fixed in 0.8.15, now also fixed
  for the browser-STT audio-analysis path.

Changed

• Added VAD state-transition diagnostic logging: speech detected (with RMS +
  capture source), silence-after-speech (with timer duration), and
  user_silence sent. Visible in the voice debug panel and /api/voice/log.

---

How to install

The canonical artifact for the harness is the container image:

docker pull ghcr.io/psycherosai/psycheros:<version>
# or `:latest` for the most recent release

See the README's Docker block
for the recommended docker run invocation, env-var reference, and volume
layout.

The archives prefixed Psycheros- below are GitHub's auto-attached
full-monorepo source — useful for building the image yourself, but the published
image at GHCR is the supported install path.

psycheros-v0.8.15

Fixed

• Voice-chat config <script> tags (sttProvider, pttEnabled, etc.) were
  orphaned during fragment mounting — only the overlay <div> reached the DOM.
  On mobile (Chrome Android) this silently fell back to browser STT instead of
  the configured Deepgram provider, producing garbled transcriptions and sending
  zero audio frames to the server. Desktop (Tauri) was unaffected. All fragment
  children now mount correctly.
• Silence detector (VAD) now starts when push-to-talk is toggled off mid-call
  for server-side STT. Previously it only ran at call init when PTT was globally
  off — so toggling PTT off after call start left the call stuck on RECORDING
  indefinitely.
• Silence detector check loop no longer dies when the voice WebSocket isn't open
  yet. An early return cancelled rescheduling, leaving VAD dead for the entire
  call until a full restart.

---

How to install

The canonical artifact for the harness is the container image:

docker pull ghcr.io/psycherosai/psycheros:<version>
# or `:latest` for the most recent release

See the README's Docker block
for the recommended docker run invocation, env-var reference, and volume
layout.

The archives prefixed Psycheros- below are GitHub's auto-attached
full-monorepo source — useful for building the image yourself, but the published
image at GHCR is the supported install path.

psycheros-v0.8.13

Fixed

• Fixed silent config fetch failure in voice chat by falling back to
  /api/voice/status when the HTML config peek returns 'browser'. This recovers
  the correct STT provider and activates native mic capture.
• Simplified context inspector label to just the turn number instead of the
  unreadable "Turn X / Y of Z" on conversations longer than 50 turns.

Changed

• Added comprehensive diagnostics for voice calls: POST /api/voice/log
  endpoint, unconditional diagnostic logs at config resolution, native capture
  success/failure, and config peek recovery.
• Added STT/TTS provider info to /api/voice/status response and voice chat
  logs.
• Added live-session capture event logging to prove frames route through the
  live voice session (not the test probe).

---

How to install

The canonical artifact for the harness is the container image:

docker pull ghcr.io/psycherosai/psycheros:<version>
# or `:latest` for the most recent release

See the README's Docker block
for the recommended docker run invocation, env-var reference, and volume
layout.

The archives prefixed Psycheros- below are GitHub's auto-attached
full-monorepo source — useful for building the image yourself, but the published
image at GHCR is the supported install path.

entity-core-v0.4.4

Fixed

• Fixed orphaned vec_graph_nodes entries that appeared after prune/merge
  phases. verifyVectorTableSync now removes orphans on startup instead of just
  warning; consolidateGraph adds a Phase 5 sweep that removes vec entries for
  any soft-deleted node.

---

How to install

This package ships as a scoped source release. Download entity-core-v*.tar.gz
(or .zip) — those archives contain the entity-core package tree only.

The archives prefixed Psycheros- are GitHub's auto-attached full-monorepo
source — they contain the whole workspace, useful if you want the companion
packages too.

psycheros-v0.8.12

Fixed

• Fixed a memory loading bug in the editor where GET handler for
  /fragments/settings/memories/:granularity/:date stripped the slug before
  calling readMemory, so entity-core fell into findMemoryByDate and returned
  the first file matching the date prefix. On dates with multiple significant
  memories, the editor loaded the wrong memory's body while showing the correct
  title.

Changed

• Added comprehensive voice pipeline diagnostic logging to trace audio flow from
  mic capture through to STT output. Five new logging points: JS
  handleVoiceMessage (daemon message types), JS channel.onmessage (frame
  send counts), daemon binary frame reception (WS handler), daemon STT
  lifecycle, and Rust tap block RMS audio level (detects silent capture). With
  voiceChatDebug on, a single voice call shows exactly where audio dies in the
  pipeline.
• Added frame-send logging to voice.js to track dropped frames when voiceWs
  isn't open yet and the first successful frame transmission to the WebSocket.
  Logging is throttled to avoid drowning the console.
• Restored the macOS mic permission check (authorizationStatus) before calling
  requestAccess to skip redundant permission prompts on subsequent voice
  calls.

---

How to install

The canonical artifact for the harness is the container image:

docker pull ghcr.io/psycherosai/psycheros:<version>
# or `:latest` for the most recent release

See the README's Docker block
for the recommended docker run invocation, env-var reference, and volume
layout.

The archives prefixed Psycheros- below are GitHub's auto-attached
full-monorepo source — useful for building the image yourself, but the published
image at GHCR is the supported install path.

launcher-v2-v0.2.42

Fixed

• macOS mic permission prompt now appears only once (first voice call) instead
  of every call. Restored the authorizationStatus check before requestAccess
  — if macOS already reports "authorized" (status 3), skip the request and
  return early without triggering the system permission prompt.

Changed

• Added RMS audio level logging to the Rust tap block (every 100th frame) to
  detect when mic capture is producing silence vs actual audio. 0.0 = dead
  silence (wrong device or muted), ~0.05 = room noise, ~0.1+ = speech. This
  catches "frames are flowing but mic isn't capturing" — the hardest issue to
  diagnose without direct measurement. Uses Arc<AtomicU32> for the frame
  counter since the tap closure must be Fn (not FnMut) and is called from
  AVAudioEngine's render thread.

---

macOS

Download Psycheros_<version>_aarch64.dmg, double-click to mount, drag the
Psycheros app into Applications/.

The build is unsigned by deliberate decision. macOS Gatekeeper will refuse
to open it on first launch with "can't be verified." The dance to bypass this is
one-time per install:

1. Right-click the Psycheros app in Applications → Open.
2. The OS shows a confirmation dialog with an Open button (vs. only
   Cancel in the default double-click flow).
3. Future launches work normally.

Once installed, the app supervises a persistent Psycheros daemon via launchd —
closing the window doesn't stop me. See the in-app Diagnostics card for paths
and the Settings card for entity config.

Psycheros.app.tar.gz is the auto-updater bundle format — not useful for
first-install; it's consumed by tauri-plugin-updater when the launcher checks
for shell-binary updates.

Windows

Download Psycheros_<version>_x64_en-US.msi and run it.

Windows SmartScreen will flag the installer as unrecognized because it is
unsigned. Click More info → Run anyway to proceed. This is one-time
per version.

The installer places Psycheros in C:\Program Files\Psycheros\. At the end of
the wizard, the launcher registers a Task Scheduler job that supervises the
daemon — closing the launcher window or the system-tray icon does not stop
the daemon. The daemon starts at login and restarts automatically on crash.

Psycheros_<version>_x64-setup.exe is the NSIS-based alternative installer with
the same content — use whichever you prefer.

launcher-v2-v0.2.41

Fixed

• macOS mic-capture now correctly resamples audio at any hardware sample rate,
  not just 48kHz. The integer decimation from 0.2.40 only handled 48kHz cleanly;
  at 24kHz (which AVAudioEngine hands out for some AirPods configurations) it
  shipped audio at the wrong rate, which would have produced garbage
  speech-to-text. Switched to linear interpolation with a float ratio — works
  for 48kHz, 44.1kHz, 32kHz, 24kHz, and 16kHz hardware formats. Also adds a
  defensive bounds check to skip empty buffers and avoid reading past the input.

---

macOS

Download Psycheros_<version>_aarch64.dmg, double-click to mount, drag the
Psycheros app into Applications/.

The build is unsigned by deliberate decision. macOS Gatekeeper will refuse
to open it on first launch with "can't be verified." The dance to bypass this is
one-time per install:

1. Right-click the Psycheros app in Applications → Open.
2. The OS shows a confirmation dialog with an Open button (vs. only
   Cancel in the default double-click flow).
3. Future launches work normally.

Once installed, the app supervises a persistent Psycheros daemon via launchd —
closing the window doesn't stop me. See the in-app Diagnostics card for paths
and the Settings card for entity config.

Psycheros.app.tar.gz is the auto-updater bundle format — not useful for
first-install; it's consumed by tauri-plugin-updater when the launcher checks
for shell-binary updates.

Windows

Download Psycheros_<version>_x64_en-US.msi and run it.

Windows SmartScreen will flag the installer as unrecognized because it is
unsigned. Click More info → Run anyway to proceed. This is one-time
per version.

The installer places Psycheros in C:\Program Files\Psycheros\. At the end of
the wizard, the launcher registers a Task Scheduler job that supervises the
daemon — closing the launcher window or the system-tray icon does not stop
the daemon. The daemon starts at login and restarts automatically on crash.

Psycheros_<version>_x64-setup.exe is the NSIS-based alternative installer with
the same content — use whichever you prefer.

launcher-v2-v0.2.40

Fixed

• macOS mic-capture no longer crashes on start. The crash log from 0.2.39
  pinpointed the failure to installTapOnBus_bufferSize_format_block — Apple
  throws NSInvalidArgumentException when the tap format doesn't match the
  input node's output format. We were requesting 16kHz Float32 on a hardware
  input node that runs at 48kHz. The fix queries the hardware's actual format
  via outputFormatForBus(0) and decimates in Rust to ~16kHz (take every Nth
  sample) — same algorithm the original JS path used. Mic capture now works on
  all Macs regardless of native sample rate.

---

macOS

Download Psycheros_<version>_aarch64.dmg, double-click to mount, drag the
Psycheros app into Applications/.

The build is unsigned by deliberate decision. macOS Gatekeeper will refuse
to open it on first launch with "can't be verified." The dance to bypass this is
one-time per install:

1. Right-click the Psycheros app in Applications → Open.
2. The OS shows a confirmation dialog with an Open button (vs. only
   Cancel in the default double-click flow).
3. Future launches work normally.

Once installed, the app supervises a persistent Psycheros daemon via launchd —
closing the window doesn't stop me. See the in-app Diagnostics card for paths
and the Settings card for entity config.

Psycheros.app.tar.gz is the auto-updater bundle format — not useful for
first-install; it's consumed by tauri-plugin-updater when the launcher checks
for shell-binary updates.

Windows

Download Psycheros_<version>_x64_en-US.msi and run it.

Windows SmartScreen will flag the installer as unrecognized because it is
unsigned. Click More info → Run anyway to proceed. This is one-time
per version.

The installer places Psycheros in C:\Program Files\Psycheros\. At the end of
the wizard, the launcher registers a Task Scheduler job that supervises the
daemon — closing the launcher window or the system-tray icon does not stop
the daemon. The daemon starts at login and restarts automatically on crash.

Psycheros_<version>_x64-setup.exe is the NSIS-based alternative installer with
the same content — use whichever you prefer.