Skip to content

Commonplace 0.2.4

Choose a tag to compare

@0xPD33 0xPD33 released this 22 Oct 21:17
· 92 commits to master since this release

Commonplace is an offline research assistant for Android. It answers from Wikipedia and other reference text on the phone and cites its sources. The app has no INTERNET permission.

Changes since 0.2.3

  • "Are you sure?" checks the answer. A doubt such as "are you sure?", "that's wrong" or "double check that" searches the last question and its answer's first claim again, and the model checks its answer against the sources. Before, the app often replied "I'm confident!" and did nothing.
  • Answers beyond the sources are marked. When the sources do not answer a question, the model says so first and can then answer from general knowledge. The app underlines those sentences: a sentence without a citation gets the underline when the evidence does not have most of its rare words.
  • Rounded numbers are not underlined. "26.5 times" matches 26.52 in the evidence, and "16.4 million km²" matches 16,431,599 km².
  • Recent opens the conversation. A question under Recent on the home screen opens its saved conversation, and "All conversations" opens the history.
  • Shorter topic chips. Chapter and page titles, such as "Microbiology (Boundless): 12.1D: Vaccine Safety", no longer appear as chips.
  • The TPU engine choice is off. LiteRT-LM cannot use the Pixel's Tensor chip yet, so Settings shows the choice greyed out and every answer uses the CPU.
  • Fixes. "Should I get one before my holiday?" no longer links the song "Should I Stay or Should I Go". A follow-up after "The sources do not…" searches the right topic. The app imports a second download such as commonplace-starter (1).tar. Diagnostics shows the git hash of the release, not "Build dev".

Install 0.2.4 over any earlier version. The installed packs stay.

Install

  1. Download commonplace-0.2.4.apk below on an Android phone (64-bit ARM, Android 12 or later) and install it.
  2. Open Commonplace and tap Open Library. Under Get more, tap the starter set marked For this phone and tap Download. The browser saves commonplace-starter.tar (11.3 GB, 12 GB phones) or commonplace-starter-small.tar (8.1 GB, 8 GB phones) from Hugging Face.
  3. Tap Install from Downloads and select the file. The app checks every byte with SHA-256 during the import.
  4. Ask a question.

You need 8 GB of RAM and free storage of about twice the starter set during the first import. docs/INSTALL.md lists the other packs and the install from a computer.

Packs

The packs did not change since 0.2.3. They are in the Hugging Face dataset 0xPD33/commonplace-packs, folder packs-2026-09.

File Size Contents
commonplace-starter.tar 11.3 GB Starter set for 12 GB phones: Wikipedia starter (2M articles), Ling-3.0-tiny, Wikidata facts
commonplace-starter-small.tar 8.1 GB Starter set for 8 GB phones: Wikipedia starter, MiniCPM5-2B, Wikidata facts
commonplace-breadth.tar 2.0 GB 13 packs: textbooks, Wikibooks, Wikivoyage, Wiktionary, DevDocs, medical and travel health pages and more
enwiki.tar 19.2 GB All of English Wikipedia (replaces the starter Wikipedia)
arxiv-abs.tar 3.3 GB arXiv abstracts
stackexchange.tar 2.5 GB Stack Exchange questions and answers from 51 sites
minicpm5-2b.tar 1.4 GB The small model alone, for a phone with the 12 GB starter set
voice-en.tar 0.3 GB Voice input, English

Verify the APK

  • SHA-256 of commonplace-0.2.4.apk: eb81c4525b620c1f16c9c07fe6984e6db9aa6d463c6939d227f8aa07401a3783
  • Signing certificate SHA-256: c17ec0b2fe898fa8eb6b172c14be053c5748965a8803bd0f021de737a65aa10b
  • Permissions: microphone, foreground service and notifications. No network permission.

Known limits

  • In a long session the phone heats up, and the first word takes longer on a hot phone.
  • The model still states some facts and numbers wrongly. The underline is a word check, not a fact check: about 1 in 4 underlined sentences is in the sources after all.
  • Long-tail, travel and fact questions score lowest, at 0.40 to 0.42 of the baseline (EVAL).
  • The app uses only the CPU. The Tensor G5 NPU does not work with the current LiteRT-LM.
  • Real speech recognition is tested on the desktop, not yet on a phone's microphone.