Skip to content

Getting started

Jump to bottom
jp edited this page Jun 28, 2026 · 8 revisions

Getting started

Install (sideload)

Candela is sideload-only today. Tagged releases attach a signed debug APK to the GitHub release. Every push to main also produces a debug APK as a CI artifact.

  1. Download the latest candela-vX.Y.Z.apk from Releases.
  2. Open it on your Android device (Android 8.0+; targetSdk = 35).
  3. Allow "Install from unknown sources" if prompted.
  4. From v0.4.15 onward the signing keystore is committed to the repo, so debug-to-debug and release-to-release upgrades land cleanly. Pre-v0.4.15 builds used per-machine debug keystores — upgrading from those needs a one-time uninstall first.

First launch

On a brand-new install, Candela first runs a short three-screen welcome (Welcome → voice picker → first fiction — see Export & extras). After that, Candela opens on the Library tab. The bottom-of-screen dock is {Playing · Library · Voices · Settings} (the four-tab dock settled in v0.5.50 after the v0.5.40 nav restructure).

Library leads with the brass TechEmpower hero card — tap it to reach TechEmpower Home (Discord peer support, dial 211, Browse the resource library, About TechEmpower). Library sub-tabs sit above the dock: Library · Browse · Follows · Inbox · History.

  • Library — your saved fictions, filtered by All / Reading / Read / Wishlist chips. The Resume button uses smart-resume — it respects your last paused/playing intent, so it won't auto-play at you when you opened the app to read.
  • Browse — twenty-five fiction backends. The Notion chip is selected by default (v0.5.48+) and shows TechEmpower's resource library (Guides, Resources, About, Donate). Other chips: Royal Road, GitHub, RSS, EPUB, PDF, Scanned text (OCR), Outline, Memory Palace, Project Gutenberg, AO3, Standard Ebooks, LibriVox, Wikipedia, Wikisource, Radio (30k+ stations), Hacker News, arXiv, PLOS, Discord, Telegram, Palace Project, Slack, Matrix, plus the Readability catch-all for arbitrary URLs.
  • Follows — synced from Royal Road if you sign in.
  • Settings — see Settings reference.

Tap a fiction to open the detail screen. Hit "Listen" or pick a chapter to start playback. The first chapter triggers a one-time voice-model download (Piper "amy" by default, ~30 MB; KittenTTS as the lightest tier is ~24 MB shared across 8 speakers).

Picking a voice

Settings → Voice library. You'll see four local engine groups plus the optional cloud backend:

  • Piper — compact (~14–30 MB), one-speaker-per-model, fast on every device. 45 voices.
  • Kokoro — multi-speaker (~330 MB total), more expressive, heavier. 53 speakers.
  • KittenTTS (v0.5.36, #119) — the lightest tier. ~24 MB shared across 8 en_US speakers. Designed for slow devices where Piper-high struggles; the "first chapter in 10 seconds" voice family.
  • Supertonic 3 (v1.2.3, #1191) — the newest neural family. 10 en_US speakers (5F/5M) from a ~139 MB shared model. A step up in naturalness for modern devices.
  • Azure HD (optional, BYOK) — studio-grade cloud voices via Azure Cognitive Services. No local download. 20 voices including the new Dragon HD generative tier. Costs you Microsoft's published rate; Candela doesn't bill you.

Tap a voice to install it (or, for Azure, to select after key + region setup — see Settings reference). Star the ones you like to surface them in the picker. The ★ Starred section floats at the top.

For the full list, see the Voice catalog reference.

Sleep timer

Now-Playing → moon icon. Pick a duration (15/30/45/60 min) or "End of chapter". The fade tail is 10 seconds — during the fade, shake the phone (three sharp shakes) to re-arm for another 15 minutes (#150). Toggle this off in Settings → Reading if your commute is bumpy. You can also auto-enable Do Not Disturb while the timer runs, and bedtime auto-sleep arms the timer automatically when your phone enters Sleep / Bedtime mode — both live in Settings → Voice & Playback (see Settings reference).

Sign-ins (optional)

  • Royal Road — Settings → Account → Royal Road → Sign in. Webview-based; once captured, your follow list syncs into the app.
  • GitHub — Settings → Account → GitHub → Sign in. OAuth Device Flow; lifts the anon 60 req/hr cap to 5,000 and unlocks "My Repos", "Starred", and "Gists" in Browse. Toggle "Enable private repos" if you want private repos as fiction sources.
  • Memory Palace — Settings → Memory Palace → host + API key. Test the connection before committing.
  • Outline — Settings → Library & Sync → Outline → server URL + API token from your Outline user settings.
  • Azure — Settings → Voice & Playback → Azure → key + region. Hit Test connection before picking an Azure voice from the library.

Adding sources

  • EPUB folder — Settings → Plugins → EPUB → pick a folder via the Storage Access Framework. Candela parses each .epub's .opf manifest into chapters.
  • RSS feed — Browse → RSS tab → paste a feed URL, or pick from the curated suggested-feeds list (managed in candela-feeds).
  • Telegram — Settings → Plugins → Telegram → bot token. Invite the bot to a public channel; the channel becomes a fiction.
  • Slack — Settings → Plugins → Slack → workspace bot token (xoxb-…). Default OFF on fresh installs — workspaces are private and bot-token onboarding is high-friction.
  • Matrix — Settings → Plugins → Matrix → homeserver + access token. Federated rooms become fictions with same-sender message coalescing.
  • Palace Project — Settings → Plugins → Palace → pick a library from the OPDS catalog. Ships non-DRM titles only in v0.5.51; LCP DRM is deferred.

Cold launch performance

v0.5.46 cut cold launch from 6.7 s to 0.8 s on a Galaxy Tab A7 Lite — the combo is R8 minification, a Baseline Profile (regenerated on every tag push by :baselineprofile), and isDebuggable=false in release builds. The numbers are baked in; you don't need to flip a toggle.

AI chat

The chat icon on the reader opens a per-fiction chat. Pick a provider in Settings → AI; Candela supports Claude direct, Anthropic Teams (OAuth), OpenAI, Vertex, Bedrock, Foundry, and Ollama. Grounding controls let you feed the AI the current sentence, the chapter, or the entire book so far.

Long-press a word in the reader to ask "Who is X?" — useful for character/place lookups deep into a long serial.

Next steps

Candela

Start here

Using Candela

Developer

Project

Clone this wiki locally