Skip to content

Home

Jump to bottom
jp edited this page Jun 26, 2026 · 130 revisions

Candela wiki

Current version: v1.1.5 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: twenty-five 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: v1.1.5 — Voices, gears, and polish — KittenTTS v0.8 voices, gzip downloads, AGP 9 migration, inbox fix, and accessibility improvements across the board.

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

Pages

What's new in v1.1.5

Voices, gears, and polish. KittenTTS v0.8 voices, gzip downloads, AGP 9 migration, inbox fix, and accessibility improvements across the board.

  • KittenTTS v0.8 voices. Upgraded Kitten nano models from v0.1 to v0.8 — clearer, more natural speech with named speaker variants. (#1113 / #1115)
  • Supertonic 3 engine scaffold. Catalog entries, sample-rate cache, and render-job wiring for the upcoming Supertonic 3 TTS engine. (#1114 / #1120)
  • Inbox unread count. New-chapter notifications now accumulate correctly across coalesced polls instead of resetting to the latest poll's count. (#1083 / #1122)
  • Voice downloads use gzip. Model files download gzip-compressed with transparent decompression, cutting bandwidth on large ONNX models. Sub-8 KB files (voices.bin, tokens.txt) skip compression overhead. (#1112 / #1116)
  • Accessibility semantics. Settings rows, chat read-aloud buttons, and tool-call cards now announce coherently via TalkBack instead of exposing fragmented or unlabeled descendants. (#491 / #1125)
  • AGP 9 new DSL. Removed android.newDsl=false and switched to AGP 9's built-in Kotlin compilation (android.builtInKotlin=true), eliminating the separate kotlin-android plugin across all 37 modules. (#923 / #1124)
  • Dependencies. Kotlin 2.4.0, Hilt/Dagger 2.60, compileSdk 37, Coil 3.5.0, Lifecycle 2.11.0, core-ktx 1.19.0, CameraX 1.6.1, sherpa-onnx 1.13.3, Commonmark 0.29.0, Gradle 9.6.1, Compose BOM latest. (#1100–#1109, #1111)

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