Repository navigation
Releases
PocketCompanion 1.0.0
PocketCompanion 1.0.0: the first public release of the offline AI avatar companion for Android.
A lip-synced AI companion you talk to by voice, text or photo. The language model, speech recognition, voice and avatar all run on your phone's CPU, with no account and no cloud.
Install
- Download PocketCompanion-1.0.0.apk below. Your browser or file manager needs "install unknown apps" allowed.
- Open the app and tap Download and start. It downloads about 5 GB of models once (use Wi-Fi). After that it works offline.
Requirements: Android 10+, 64-bit ARM, 8 GB RAM minimum (12 GB recommended), about 6 GB free storage.
What's in it
- Voice chat with live transcription (NVIDIA Nemotron Speech Streaming 0.6B)
- Gemma 4 E2B (abliterated, Q4_K_M) through llama.cpp, with photo understanding
- 14 Kokoro voices and 11 Duix avatars with lip sync and word-by-word subtitles
- Editable instructions (system prompt) in Settings
New compared to the private development builds
- Asks before the one-time ~5 GB download and checks free space
- Speech recognition model downloads automatically
- Photos are processed at 512 px: same description quality, answers about twice as fast
- Six English voices (American and British), new icon, credits in Settings
Full history: CHANGELOG.md · Demos: https://solvai.github.io/PocketCompanion/
Verify
- APK SHA-256:
fca74a6e7d55f478e3d5b5b72149f81572b3d4263e1b5b669296ebb1888b69fa - Signing certificate SHA-256:
05:FB:14:D4:0E:DD:9B:13:1C:49:5D:BB:DF:B1:87:7C:EB:EA:E0:FE:62:35:08:CF:C4:E8:1C:91:B4:49:FD:41
The default model is abliterated (refusals removed), so it answers more freely than a filtered assistant. Powered by Duix.com: the avatar engine is under the duix.com Community License, see THIRD_PARTY_NOTICES.md.
Speech model (Nemotron Speech Streaming 0.6B, 560 ms)
Speech recognition model used by PocketCompanion, downloaded by the app on first start. Not an app release.
Model: NVIDIA Nemotron Speech Streaming en 0.6B, 560 ms chunks, int8, as exported for sherpa-onnx.
- Original model: https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b
- sherpa-onnx export:
sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25.tar.bz2from https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models - These four files are that archive's contents, unmodified. They are mirrored here because the app downloads single files, and the upstream release only has the .tar.bz2.
Licensed by NVIDIA Corporation under the NVIDIA Open Model License.
SHA-256:
7d932213491ad355c6e5576705dc3494731a52af87d7a1b954559340147909d8 encoder.int8.onnx
0be9702c2f427a2b6bb241d298e0d3836a558de1f5b9fd3018f1cce6e2b3fa98 decoder.int8.onnx
a35eac38a22ebceb04d230ed7afe0d68f446ba6914a036b97f14fece95967e23 joiner.int8.onnx
dc0b4584ab2e4ddbf888425c076c61b736e7356a015250db7d307e6f1a8188ff tokens.txt