Skip to content

Home

Jump to bottom
github-actions[bot] edited this page Jul 10, 2026 · 130 revisions

Candela wiki

Current version: v1.13.0 Candela is TechEmpower's accessible resource app. Browse free tech guides, connect with our peer-support Discord, dial 211 for local help — and listen to any of it through a neural-voice audiobook engine that reads everything aloud.

Under the hood: thirty-three fiction backends (Royal Road, GitHub, RSS / Atom feeds, local EPUB folders, local PDF files, on-device OCR, self-hosted Outline wikis, Memory Palace, Project Gutenberg, LibriVox, Archive of Our Own, Standard Ebooks, Wikipedia, Wikisource, Radio with 30k+ stations, Notion, Hacker News, arXiv, PLOS, Discord, Telegram, Palace Project, Slack, Matrix, Readability web articles); three in-process voice families (Piper, Kokoro, KittenTTS) with optional Azure HD cloud voices; AI chat per fiction across seven LLM providers with cross-fiction memory, function calling, and multi-modal image input; built for Android phones, tablets, and Wear OS.

📦 Latest release: v1.13.0 — Fabled Jewel — Voice Notes arrives — record a thought, meeting, or memo and Candela transcribes it on-device and, with your consent, turns it into an AI summary, saved as a searchable note. Plus three Settings additions.

📖 Source: github.com/techempower-org/candela · License: GPL-3.0 · Site: candela.techempower.org

Pages

What's new in v1.13.0

Fabled Jewel. Voice Notes arrives — record a thought, meeting, or memo and Candela transcribes it on-device and, with your consent, turns it into an AI summary, saved as a searchable note. Plus three Settings additions.

  • Voice Notes (#1657) — a Plaud-style capture surface: record audio → on-device transcription (Whisper) → optional, consent-gated AI summary (title + key points / action items) → saved as a searchable note; typed notes are first-class too. Privacy-first by construction: audio stays on the device and is excluded from cloud backup + device-transfer; transcription runs entirely on-device; and the transcript leaves the phone only when you explicitly tap Summarize with an AI provider configured. Reached from the Notes pill on the tablet rail, or the waveform action in the Library top bar on phones.
  • Sleep-timer default duration (#1590) — pick your preferred default in Voice & Playback instead of the fixed 15 minutes.
  • Palace Project library config (#1591) — surfaced in the Content Sources settings.
  • AO3 account row (#1592) — sign-in/session state and sign-out in Account settings.
  • Voice Notes lives in a separate notes.db (cleanly backup-excludable), transcribes via a durable WorkManager + foreground-service job that survives backgrounding, and uses column-scoped DAO writes so a live transcription can't clobber a concurrent edit (#1663). Built across six phases with adversarial review at each step; note content is never logged.

Recent v0.5 highlights

  • Cold launch 6.7 s → 0.8 s on Galaxy Tab A7 Lite (v0.5.46 — R8 + Baseline Profile + isDebuggable=false).
  • Full PCM cache series — streaming-tee, cache-hit playback, background pre-render, Settings UI, status icons (v0.5.47–v0.5.49).
  • Twelve a11y findings closed (v0.5.43) — high-contrast brass-on-near-black theme, reduced-motion fold-in, TalkBack pacing.
  • Nav settled — four-tab dock {Playing · Library · Voices · Settings} (v0.5.50 final, after v0.5.40 restructure).
  • Cross-device InstantDB sync (v0.5.39+) — library, follows, positions, bookmarks, pronunciation, encrypted secrets.
  • KittenTTS lightest-tier voice family (v0.5.36 — #119).

Quick links

Candela

Start here

Using Candela

Developer

Project

Clone this wiki locally