Skip to content

Changelog ​

All notable changes to Conjure, organized by development phase.

INFO

Conjure is in beta. Version numbers below refer to internal development phases, not public releases.

Phase 12 — Audio Preprocessing, Agent Permissions & TTS Latest ​

  • Audio preprocessing pipeline — per-frame noise gate, 80Hz high-pass filter, RMS normalization applied in Rust sidecar before whisper inference
  • Tuned whisper params — no_speech_thold(0.4), entropy_thold(2.4) reduce hallucinations
  • Skip tool permissions toggle in Agent mode settings
  • TTS pill overlay — purple pill with speaker icon during playback, click to stop
  • TTS speak-selected-text — clipboard fallback, toast feedback on failures
  • Settings reorganization — dissolved MoreSettingsDialog into 5 logical sections (Appearance, Input, Processing, Account, Data & Privacy)
  • Boring Avatars — generated SVG avatars with regenerate button
  • Cost estimation — shared cost utils, aggregate cost cards on Analytics and Home
  • Contextual hints — dismissable tips on 7 pages for feature discovery
  • Voice commands reference dialog — in-app reference for all 36 voice commands
  • Help & Support menu — shortcuts, voice commands, about links
  • Dialog fixes — sticky footer, overflow-x-hidden, missing dialogs mounted
  • UI audit fixes — AudioWaveform color, ListTile keyboard handler, sidebar rem scaling, conversation menu fix
  • Documentation site — VitePress-powered docs with 15 pages, 3D crystal hero, callouts

Phase 11 — Upstream Sync v0.0.483 ​

  • Streaming verbatim mode — real-time text output as you speak (~70ms latency)
  • Streaming text injection — type_text command for direct Unicode injection, safe for terminals and push-to-talk
  • Metal GPU acceleration — platform-specific feature flags for macOS Metal
  • Tone improvements — better pronoun preservation in Polished, Email, and Chat prompts
  • Aborted session guard — prevents proceeding with cleared session

Phase 10 — Auto-Learn Dictionary & Correction System ​

  • Auto-learn engine — tracks correction patterns, surfaces suggestions after threshold
  • Dictionary suggestions — banner with staggered animations, optimistic dismiss
  • Dictionary enhancements — search bar, type filter, delete confirmation, categories
  • Inline word correction — click-to-select words in transcript history
  • Correction dialog — dedicated view with diff segment highlighting
  • Analytics integration — correction stats card, top corrections chart
  • Nav badge — pending suggestion count on Dictionary

Phase 9 — User Profile & Data Portability ​

  • Profile dialog — avatar, bio/company/title, usage stats
  • Data export/import — selective ZIP export with per-category checkboxes
  • Security hardening — column validation, ZIP path traversal protection
  • Theme toggle — light/dark/system in Settings

Phase 8 — API Usage Tracking & Model Info ​

  • Usage tracking — per-transcription token counts, per-key aggregation
  • Cost estimation — provider-specific pricing with free tier detection
  • Model info dialog — context-aware recommendations per provider
  • Usage stats toggle — configurable in Settings

Phase 7 — Branding & Logo System ​

  • Custom logo — faceted dark crystal with spectrum ring
  • Three-tier icon system — Full, Small, Tiny variants with dark/light modes
  • Icon generation — automated PNG rendering to 93+ target locations

Phase 6 — Upstream Sync v0.0.461 ​

  • Bug fixes — CPU sidecar health check, Deepgram transcript cutoff, settings reversion
  • Features — edit API keys, double-tap Esc to discard, toast action system
  • Locale updates — all supported languages refreshed

Phase 5 — Local Processing Parity ​

  • 300s segments — local sidecar matches cloud mode duration
  • Usage stats — word count and token tracking per transcription

Phase 4 — GPU Sidecar Architecture ​

  • whisper.cpp sidecar — standalone axum HTTP server for transcription
  • Dual binaries — CPU and GPU sidecars bundled with the app
  • Runtime GPU detection — automatic CPU fallback
  • Streaming sessions — HTTP-based chunked audio upload

Phase 3 — Firebase Removal ​

  • PocketBase migration — auth flow, token refresh, local-only mode default
  • Firebase SDK removed — no more Google Cloud dependencies

Phase 2 — PocketBase Server ​

  • Server setup — collections, hooks, scripts, schema import
  • Local-only mode — works without any server

Phase 1 — Rebrand ​

  • Conjure identity — global rename from Voquill
  • New bundle IDs — com.conjure.desktop
  • Upstream remote — preserved for pulling improvements

Released under the AGPLv3 License.