Skip to content

v0.1.0 — Recorder, Replay & Voice-Over

Latest

Choose a tag to compare

@SurrealMindAI SurrealMindAI released this 28 Jun 11:19

First feature release of the greenfield frontprompt (FrontCom overlay + Python/anyio daemon).

Features

  • Recorder — toolbar-startable session recorder over the shared Picks/Regions/Events/Relations, floating overlay toolbar (stop/draw-region/picker), persistent recordings list (name/description) + per-recording timeline, survives cross-origin navigation (Python-SSoT), read-side recordings list/get via MCP + CLI.
  • Replay + Assertions + Agent-Control — replay a recording as a sequence of existing IpcRequests/page-controller actions, AssertionEntry checkpoints (selector/text/visible/url) with a pass/fail replay report, parametrized replay, agent write-surface (start/stop/replay via MCP + CLI).
  • Voice-Over — backend microphone capture coupled to the recording lifecycle, async transcription via a pluggable TranscriptionBackend (local mlx-whisper on Apple Silicon, on-the-fly model download from the overlay), selectable model index (incl. a German model), Settings tab (mic + transcription backend + model), transcript segments injected into the timeline by timestamp.
  • Timeline UX — expandable per-entry detail views resolving pick/region/relation references to real details.

Quality

  • Test coverage: backend 90.5%, frontend 96% lines (both gated, fail_under=90 / vitest thresholds).
  • Full real-Chromium e2e suite (record → assert → replay → report; voice-over with mocked STT) + codegen drift gate, all green.

Notes

  • Schema versions: state 0.11.0, bridge 0.11.0, IPC 0.8.0, overlay 0.11.0.
  • mlx-whisper ships as a platform-marked core dependency (Apple Silicon only); sounddevice/huggingface_hub are core.