Skip to content

content: TobiBoard runs its AI on-device, and it was never submitted to IzzyOnDroid - #51

Merged
leinss merged 1 commit into
releasefrom
feature/tobiboard-content
Aug 26, 2026
Merged

content: TobiBoard runs its AI on-device, and it was never submitted to IzzyOnDroid#51
leinss merged 1 commit into
releasefrom
feature/tobiboard-content

Conversation

@leinss

@leinss leinss commented Aug 26, 2026

Copy link
Copy Markdown
Owner

Two problems with what the site says about TobiBoard.

The AI features are described as cloud-only

Every surface said the voice-to-text and text fix "only run when you add your own
OpenRouter or PayPerQ key". They have defaulted to the on-device engines for a
while: PREF_AI_PROVIDER defaults to local, dictation runs Parakeet TDT 0.6B
through sherpa-onnx, and text fix runs Qwen 2.5 through MediaPipe. No account, no
key, works in aeroplane mode. Cloud providers are the opt-in alternative, not the
only path.

Reframed across all seven surfaces so the site is self-consistent, and stated the
cost plainly rather than only the upside: each model is a one-time download of a
few hundred megabytes, and a phone is slower than a datacentre.

A claim that was never true

The project page said "IzzyOnDroid: inclusion request submitted, in review". It
was never submitted and the app is not eligible: the release APK is roughly four
times their per-app size limit, and the bundled speech and LLM runtimes are
prebuilt binaries they do not accept. Removed, and replaced with a short note on
why the app ships outside F-Droid's main repository.

Files

  • src/content/projects/TobiBoard/index.md — rewritten AI section, IzzyOnDroid
    line removed, clipboard manager mentioned, frontmatter description updated
  • src/content/blog/{en,de}/tobiboard-vs-gboard/index.md — comparison row, AI
    section and conclusion, both languages
  • src/consts.ts, public/llms.txt — app tagline
  • src/content/staticPages/now.md, uses.md — the "bring-your-own-key for
    everything" framing, plus the on-device LLM line

Google Play is deliberately not mentioned: only the internal test track is live.

Test plan

  • pnpm build clean, 82 pages, 1758 internal links checked, none broken
  • pnpm lint clean

…to IzzyOnDroid

The site described both AI features as bring-your-own-key only. They have
defaulted to the on-device engines for a while: Parakeet TDT 0.6B through
sherpa-onnx for dictation, Qwen 2.5 through MediaPipe for text fix, with cloud
providers as the opt-in alternative. Reframed across the project page, both
language versions of the Gboard comparison, the app tagline, llms.txt, /now and
/uses, and noted the honest cost (a few hundred megabytes per model, and a phone
is slower than a datacentre).

Also removed the project page's claim that an IzzyOnDroid inclusion request was
submitted and in review. It was never submitted: the release APK is roughly four
times their per-app size limit and the bundled speech and LLM runtimes are
prebuilt binaries they do not accept. Replaced with a note on why the app ships
outside F-Droid's main repository.
@leinss
leinss merged commit e1a9800 into release Aug 26, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant