Satori – Offline Android voice notes with on-device STT and local LLMs

An independent developer released Satori, an open-source Android voice-notes app that runs every stage of its pipeline 100% offline on the phone, using NVIDIA Parakeet v3/v2, Cohere Transcribe and AI4Bharat IndicConformer speech models via sherpa-onnx and ONNX Runtime 1.28.2, plus Gemma 3 1B, Gemma 4 E2B and Qwen 3.5 0.8B for note synthesis on custom-patched llama.cpp ARM64 kernels. Satori performs live transcription, first-person note synthesis, to-do extraction and mind-map linking with zero accounts, zero cloud servers and zero telemetry, touching the network only to download model weights from Hugging Face, and adds English↔German live translation and a Malayalam→English reading pass via AI4Bharat IndicTrans2 200M. The app supports the Motorola razr 70 plus cover screen (1080×1272, 480×410 dp), including power-key double-press capture over the lock screen with layout clearance around the 232 px camera band.

Fully offline voice recorder, live transcriber, journal & meeting synthesizer, and personal mind map for Android — powered 100% by on-device AI. I wanted a way to capture thoughts on the go with a quick double-press of a physical hardware key—without unlocking my phone or getting pulled into a screen—and have those notes accumulate over time into a living mind map. I use Satori every day for journaling , quick thoughts , ideas , self-development , local AI and engineering notes , and transcribing meetings . I didn't have the time to audition a dozen different note-taking apps, and I refused to pay recurring subscriptions just to hand my private voice recordings and personal reflections over to someone else's cloud servers. Instead, I built this. Every stage of Satori runs 100% offline on the phone : - Live speech recognition while you walk or sit in a meeting NVIDIA Parakeet v3/v2 , Cohere Transcribe , AI4Bharat IndicConformer , and Silero VAD via sherpa-onnx and ONNX Runtime 1.28.2 . - First-person note synthesis, session naming, to-do extraction, and mind-map linking via local instruction-tuned language models Gemma 3 1B , Gemma 4 E2B , Qwen 3.5 0.8B running on custom-patched llama.cpp ARM64 kernels. - Neural reading & translation bridges for live English ↔ German conversation translation and an internal Malayalam → English reading pass AI4Bharat IndicTrans2 200M so Malayalam voice notes receive the same structured English notes, reminders, and mind-map connections. - Zero accounts, zero cloud servers, zero telemetry. The network is touched only when downloading model weights directly from Hugging Face. I am sharing Satori as open source hoping it will be useful to anyone who wants to clone, customize, and build a private voice companion for themselves or for the people they love. Built around the Grille typographic design system Space Grotesk + Doto dot-matrix tabular counters + D71921 recording red on 000000 OLED black , Satori is engineered for both standard Android displays 1080×2640 and the Motorola razr 70 plus external cover screen 1080×1272 , 480×410 dp , including hardware power-key double-press capture over the lock screen and automatic layout clearance around the 232 px physical camera band. Double-pressing the physical power key wakes the folded phone straight into recording QuickKeyActivity , keeping all interactive controls clear of the 232 px hardware camera and flash band in both standard ROTATION 0 and tent/flipped ROTATION 180 postures. | Cover View | Use Case & What It Shows | |---|---| | Quick Power-Key Capture cover-recording.png | Capturing an evening walk reflection without unfolding the phone: Doto dot-matrix timer, Voice note category selector, EN language key, streaming Parakeet v3 transcript, and the tactile Grille Stop button positioned above the camera band. | | Cover Home cover-home.png | Browsing recent Journal , Voice note , and Meeting sessions on the folded display with tabular Doto durations, search/mind-map/settings actions, and one-tap Grille recording. | | Live Translation cover-translate.png | Real-time German → English translation Record as → Translate , pairing the spoken German transcript with live English translations on the cover display. | | Cover Mind Map cover-mindmap.png | Interactive radial knowledge graph centered on You 10 recordings across Journaling , Local LLMs & AI , Self-dev , Quick Ideas , Meetings , Deep Work , and Habits , adapted for the square cover viewport. | Each recording belongs to a category— Voice note , Meeting , Idea , To-do , or custom categories like Journal —that governs how the on-device LLM structures your notes and whether spoken action items become scheduled reminders or ideas. | Inner View | Use Case & What It Shows | |---|---| | Home inner-home.png | Open To do list Block 2 hours for deep work before 10 AM , 1 suggested inbox link, 1 idea link, chronological Recordings across Journal , Idea , Meeting , and Voice note , and the docked Grille record bar. | | Active Recording inner-recording.png | Real-time streaming transcription Parakeet v3 with live partial utterance tracking and instant Record as / language switching. | | Record as Sheet inner-record-as.png | Switching category Voice note , Meeting , Idea , To-do , Translate or spoken language English or German , Malayalam before or during a recording. | | Categories Manager inner-categories.png | Built-in categories plus custom user categories Journal — Daily reflections, habits and self-dev notes , each mapped to a note-synthesis style Short notes and to-dos , Decisions, facts and to-dos , or Sections for a long meeting . | | Journaling & Self-Dev inner-journal.png | First-person journal synthesis Evening journal: deep work and focus distilled from an evening walk reflection, with expandable transcript provenance Show what was said . | | Technical & LLM Notes inner-details.png | Structured engineering notes On-device LLM latency and KV cache capturing concrete numbers and trade-offs in first-person voice without generic AI filler. | | Meeting Transcript & Player inner-transcript.png | Timestamped meeting transcript Architecture sync: local LLM latency with the built-in audio player Play from here on any line and file size readout. | | To-Dos & Suggested Inbox inner-todos-suggested.png | Calendar-grounded open To do reminders alongside the Suggested to-dos review sheet Keep , Dismiss , Move to ideas , Keep all for tasks inferred from meetings and journal reflections. | Over days and weeks, Satori connects your recordings into a two-level knowledge graph Themes and per-recording Topics and gathers spoken wishes into a dedicated Ideas collection. | Inner View | Use Case & What It Shows | |---|---| | Mind Map Overview inner-mindmap.png | Ego-centric knowledge graph centered on You Core , Recent , All filters, theme search, and recording counts across Journaling , Local LLMs & AI , Self-dev , Quick Ideas , Meetings , Deep Work , Streaming STT , and Habits . | | Theme Node Drill-Down inner-mindmap-node.png | Inside the Journaling theme node 4 recordings · Pinned in Core : radial sub-graph of child topics Deep work , Morning walk , Evening review , Attention , Sleep & energy , Weekly reset , Reading habit , Stoic notes , AI-written Summary across recordings , and linked Topics in this theme . | | Ideas Collection inner-ideas.png | Every spoken wish or idea extracted across your Journal and Meeting recordings, grouped by source session and convertible into a scheduled to-do in one tap. | | On-Device Model Manager inner-models.png | Local model manager Parakeet v3 , Parakeet v2 , Cohere Transcribe , Malayalam AI4Bharat , Better transcript after Stop , and custom recognition vocabulary . | - 100% Offline & Private : Every byte of audio, every transcript segment, every synthesized note, and your entire mind map live in a local Room SQLite database AppDatabase v13 on your phone. Nothing is ever uploaded to a server. - "The Recorder Wins" Preemption : Starting a recording—from the Home screen, the cover screen, or a double-press of the physical power key over the lock screen—opens the microphone immediately and preempts any running background AI task within milliseconds. - First-Person Notes, Zero AI Slop : Generic meeting bots write third-person filler "The speaker discusses..." . Satori's prompts and GBNF grammar schemas force the local LLM to write direct, first-person notes "Pre-faulting GGUF weights at 120 MB/s prevents audio underruns" , verified by PromptEcho filters. - Calendar-Accurate To-Dos without LLM Math : Small language models hallucinate calendar arithmetic. Satori spots relative dates and times deterministically TodoTimeResolver for English/German and MalayalamTodoCues for Malayalam relative to the recording's start timestamp, passing verified T1 / T2 tokens to the model and grounding every extracted item against the cited segment TodoGrounding . - Engineered for Real Phone Memory Limits : Multi-gigabyte GGUF models are pre-faulted at a paced 120 MB/s 0 MB/s during active recording with MAP POPULATE disabled, /proc/meminfo MemAvailable safety floors, and immediate page-cache reclamation madvise MADV PAGEOUT on unload. - Zero-wait microphone capture : RecordingService opens AudioRecord 16 kHz mono PCM the instant you tap Record , buffering audio while the speech recognizer loads in the background so the first words you speak are never lost. - Categories & Record as : - Ships with four starter categories— Voice note , Meeting , Idea , and To-do —plus custom user-defined categories Settings → Categories , such as Journal . Each category maps to a recording style SIMPLE LISTENING , SHORT MEETING , or LONG MEETING that governs how its notes and action items are structured. - Switch categories before recording, mid-recording via the top Record as chip, or retroactively on any finished recording via Change type which rewrites the notes in the new category's style while preserving your title and to-dos . - Ships with four starter categories— - Spoken languages & mid-recording switching : - English and German out of the box via NVIDIA Parakeet TDT 0.6B v3 ~670 MB . - Malayalam മലയാളം via AI4Bharat IndicConformer 120M INT8 ~140 MB . When Language in Record as is enabled, you can switch between English or German and Malayalam mid-recording between two sentences. - Custom vocabulary Words to recognize : Add names, acronyms, and domain terms one per line in Advanced settings to boost their recognition in Parakeet v3 during live capture and Transcribe again . - Double-press power key QuickKeyActivity : Binds to Motorola's hardware double-tap power gesture com.motorola.mykey , debouncing accidental bounce presses within 1,000 ms QuickKeyRepeatFilter and waking over the lock screen Hold to stop , Hold to change for pocket safety . - Selecting Translate in Record as REAL TIME TRANSLATION transcribes spoken English or German and translates finished sentences into the target language configured under Settings → Translate into English or German . - Automatically detects the spoken language German into English , or Already English: nothing to translate when input and target match . - Displays a dedicated Translation tab, running either live every few sentences As you speak or in one batch after stopping After recording . When you press Stop , the microphone closes and RecordingFinisher processes finished recordings sequentially in LlmProcessingService while leaving the recorder free for your next thought: php flowchart LR Stop "Stop Recording" -- Lines "1. Lines\n Flush streaming STT " Lines -- Redo{"2. Better transcript\nafter Stop?"} Redo -- "Parakeet v2 / Cohere /\nIndicConformer" -- Retranscribe "Re-transcribe WAV\n Making a better transcript " Redo -- "Off / Skipped" -- IsMl{"3. Malayalam\nrecording?"} Retranscribe -- IsMl IsMl -- "Yes" -- Reader "IndicTrans2 200M Reader\n Malayalam - segment readings " IsMl -- "No EN / DE " -- Notes "4. Final Notes\n Gemma 4 E2B or Gemma 3 1B " Reader -- Notes Notes -- Name "5. Auto-Name\n 2-5 word title " Name -- Todos "6. To-Dos & Ideas\n Time resolver + LLM + Grounding " Todos -- Map "7. Mind Map\n Topics - Themes & Summaries " Compact 1B language models Gemma 3 1B struggle to parse Malayalam script directly, while loading a 3.3 GB model Gemma 4 E2B on every voice note is too heavy when phone RAM is tight. Satori solves this with a dedicated internal reading stage : - Visible transcript stays in Malayalam : Your recording's Transcript tab displays the Malayalam script മലയാളം produced by IndicConformer . - Under-the-hood IndicTrans2 bridge : Between Lines and Notes , :feature-reader :lib-reader-onnx runs AI4Bharat IndicTrans2 Distilled 200M INT8 242 MB , 3 ONNX graphs on the bundled ONNX Runtime 1.28.2 + SentencePiece model.SRC / model.TGT + IndicNLP normalization and Malayalam-to-Devanagari offset transliteration, beam size 5 to write internal English readings into Room segment readings , DB v13 . - Unified downstream synthesis : Notes , Auto-naming , To-dos & Ideas , and Mind Map topics read segment readings , giving Malayalam recordings the same English notes, date resolution, and grounding quality as English recordings. - Idle backfill ReadingBackfill : Malayalam recordings that do not yet have segment readings are quietly upgraded when the phone is idle, without overwriting any title, note, or to-do edited by the owner. - Open To-Dos vs. Suggested Inbox TodoCapture : - Explicit reminders explicit , first-person promises commitment , and recordings made in the To-do category go straight to your To do list on Home TodoStatus.OPEN and schedule an exact Android alarm AndroidReminderAlarm , with Done and Snooze 1 hr notification actions . - Tasks asked of you by someone else asked , implied tasks inferred , and appointment items land in the N suggested review sheet TodoSuggestionsSheet where you can Keep , Dismiss , Move to ideas , or Keep all . - Explicit reminders - Ideas IdeasScreen : - Wishes or suggestions mentioned in a recording "there should be an option to..." , "it would be nice if..." and recordings made in the Idea category are collected on the Ideas page %d ideas, %d new on Home . - Ideas link back to their source recording and segment, outlast deleted recordings Without a recording , and convert into scheduled to-dos in one tap Make it a to-do . - Wishes or suggestions mentioned in a recording - Two-level knowledge graph centered on You : - Themes MindNodeKind.THEME : Recurring subjects across your recordings e.g., Journaling , Local LLMs & AI , Deep Work . When multiple recordings link to a theme, RefreshThemeSummariesUseCase writes a 2–4 sentence cross-recording summary of what is known and decided newest recording wins when facts change . - Topics MindNodeKind.TOPIC : What an individual recording said about a theme 1–4 grounded words + up to 3 bullet points , linking directly back to the recording. - Interactive Canvas & Accessible List : Toggle between the 2D radial graph Show as map and a structured list Show as list , filter by Core pinned + top themes , Recent last 7 days , or All , and batch-index older recordings via Map N older recordings . - Audio player & seek-to-line : Play/pause, scrub bar, file size & Compressed badge, and tap any transcript line Play from here to jump playback to that segment's timestamp. - Transcribe again : Re-run speech recognition on any saved recording's audio with any installed model Parakeet v3 , Parakeet v2 , Cohere Transcribe , or Malayalam AI4Bharat . - Automatic AAC compression & retention AudioRetentionJob : Configure Compress older audio After N days , shrinking ~2 MB/min WAV files to ~0.25 MB/min AAC/M4A while keeping them playable and re-transcribable and Delete older audio After N days , removing audio while keeping transcripts, notes, to-dos, ideas, and mind-map links . Per-recording Keep this audio exempts important recordings. - Offline GPS place naming places.bin : Optional Save the place resolves a recording's starting GPS fix offline against a bundled GeoNames KD-tree app/src/main/assets/places.bin , e.g., Berlin or Near Berlin with zero network requests. Models are downloaded on demand inside the app Settings → Models and storage and Settings → Advanced and verified by byte size and SHA-256 ModelChecksums before use. | Model | Role | Languages | Size | Runtime | |---|---|---|---|---| | Silero VAD silero vad.onnx | Voice activity detection bundled in APK assets; tunable pause length, max phrase, and threshold | Language-agnostic | 1.8 MB | ONNX Runtime 1.28.2 | | Parakeet v3 PARAKEET V3 | Default live streaming speech recognizer & Transcribe again supports custom hotwords via bpe.vocab | English & German multilingual TDT 0.6B v3 INT8 | ~670 MB | sherpa-onnx | | Parakeet v2 PARAKEET V2 EN | Optional Better transcript after Stop & Transcribe again for lower English WER | English | ~660 MB | sherpa-onnx | | Cohere Transcribe COHERE TRANSCRIBE | Optional high-accuracy Better transcript after Stop & Transcribe again | English & German | ~2.9 GB | sherpa-onnx | | Malayalam AI4Bharat INDICCONFORMER ML | Live streaming, post-stop second pass, and Transcribe again for Malayalam speech IndicConformer 120M INT8 | Malayalam ml | ~140 MB | sherpa-onnx | | Model | Role | Direction | Size | Runtime | |---|---|---|---|---| | IndicTrans2 Distilled 200M INT8 | Internal post-stop reading stage segment readings so downstream LLM steps read English instead of raw Malayalam script | Malayalam → English mal Mlym → eng Latn | ~242 MB | ONNX Runtime 1.28.2 C API + SentencePiece | | Model | Setting Name | Role | Size | Quantization | |---|---|---|---|---| | Gemma 3 1B IT | Standard default | Live notes, post-stop notes, auto-naming, to-dos & ideas, live/post-stop EN↔DE translation, and mind map | ~1.0 GB | Q8 0 GGUF | | Gemma 3 1B IT | Efficient | Lower-memory alternative for live and post-stop tasks | ~650 MB | IQ4 NL GGUF | | Qwen 3.5 0.8B | Qwen 3.5 0.8B | Compact alternative model in Advanced settings | ~870 MB | Q8 0 GGUF | | Gemma 4 E2B IT | Gemma 4 E2B | Optional higher-capacity model dedicated to AI for finished recordings post-stop notes, naming, to-dos, and mind map when free RAM permits | ~3.3 GB | Q4 0 QAT GGUF | - Paced Pre-Faulting ModelLoadPacing : Custom patches on llama.cpp 0001-load-progress-bytes.patch , 0002-map-populate-switch.patch , 0003-mapped-hook.patch disable MAP POPULATE and pre-fault GGUF pages at a capped 120 MB/s 0 MB/s while a recording is active so disk reads never starve the UI thread or audio capture buffer. - Page-Cache & Arena Reclamation : Releasing a model calls madvise MADV PAGEOUT / posix fadvise POSIX FADV DONTNEED on mapped weight pages and purges native allocator arenas. - Live /proc/meminfo Guards MemoryFloor , AnswerFloor , ParseRoom : Before loading Cohere Transcribe , Gemma 4 E2B , or IndicTrans2 , Satori checks MemAvailable against model-specific headroom floors and falls back gracefully e.g., from Gemma 4 E2B to Gemma 3 1B when memory is tight. - Disk-Cached System Prompts forgetPromptCache : All note, to-do, and mind-map system prompts are static and placeholder-free strings grille ai.xml , strings grille todo ai.xml , strings grille mindmap ai.xml , allowing :lib-llama-android to serialize their evaluated KV state to disk once and reload it across sessions without re-evaluating the system prompt on every recording. Satori is organized as a 10-module Gradle project with strict Clean Architecture boundaries :domain is pure Kotlin with zero Android dependencies : flowchart TD app ":app\nCompose Grille UI, Cover Screen, QuickKey, Foreground Services" presentation ":presentation\nViewModels, RecordingFinisher, HomeStrip, MindMap, Audio/Todo/Idea State" domain ":domain\nUse Cases, Entities, Time Resolver, Malayalam Reading Prep, Grounding" data ":data\nRoom DB v13 , Model Downloads, Memory Pacing, Offline Places, Retention" feature stt ":feature-stt\nStreaming & Offline STT + Silero VAD" feature llm ":feature-llm\nGGUF Inference, Prompt Cache & Grammar Constrained Output" feature reader ":feature-reader\nIndicTrans2 200M Malayalam-to-English Reader" lib sherpa ":lib-sherpa-onnx\nJNI Bridge to sherpa-onnx + ONNX Runtime 1.28.2" lib llama ":lib-llama-android\nCMake / JNI Bridge to llama.cpp ARMv8.0 - ARMv9.2 " lib reader ":lib-reader-onnx\nCMake / JNI Bridge to ORT C API + SentencePiece" app -- presentation app -- domain app -- data app -- feature stt app -- feature llm app -- feature reader presentation -- domain data -- domain feature stt -- domain feature stt -- data feature stt -- lib sherpa feature llm -- domain feature llm -- data feature llm -- lib llama feature reader -- domain feature reader -- data feature reader -- lib reader lib reader -- lib sherpa | Module | Responsibility | |---|---| | :app https://github.com/TinkeringDao/satori/blob/main/app | Jetpack Compose UI Grille , MainActivity singleTask cover/inner handoff , QuickKeyActivity , RecordingService , LlmProcessingService , and Hilt DI wiring. | | :presentation https://github.com/TinkeringDao/satori/blob/main/presentation | MainViewModel , RecordingFinisher , SessionsViewModel , SessionDetailsViewModel , RecordingAudioViewModel , MindMapViewModel , IdeasViewModel , SettingsViewModel , and FirstRunCoordinator state flows. | | :domain https://github.com/TinkeringDao/satori/blob/main/domain | Pure Kotlin business logic: FinishRecordingUseCase , ExtractTodosUseCase , TodoTimeResolver , TodoGrounding , MapRecordingUseCase , RefreshThemeSummariesUseCase , com.satori.domain.reading Malayalam preprocessing, and repository interfaces. | | :data https://github.com/TinkeringDao/satori/blob/main/data | Room database AppDatabase v13: sessions, segments, segment readings , insights, to-dos, ideas, mind nodes/edges, categories , resumable Hugging Face downloader, ModelLoadPacing , AudioRetentionJob WAV→M4A compression & cleanup , and offline places.bin geocoder. | | :feature-stt https://github.com/TinkeringDao/satori/blob/main/feature-stt | SherpaOnnxDataSource managing Silero VAD segmentation, streaming Parakeet v3 / IndicConformer recognition, custom hotword boosting, and offline re-transcription Parakeet v2 , Cohere Transcribe . | | :feature-reader https://github.com/TinkeringDao/satori/blob/main/feature-reader | OnnxReaderDataSource driving the 3-session IndicTrans2 200M beam-search translator for Malayalam recordings. | | :feature-llm https://github.com/TinkeringDao/satori/blob/main/feature-llm | LlamaCppDataSource managing GGUF loading, system-prompt KV disk cache, GBNF grammar enforcement, and token streaming. | | :lib-sherpa-onnx https://github.com/TinkeringDao/satori/blob/main/lib-sherpa-onnx | Prebuilt arm64-v8a libsherpa-onnx-jni.so and libonnxruntime.so 1.28.2 with Kotlin/JNI bindings. | | :lib-reader-onnx https://github.com/TinkeringDao/satori/blob/main/lib-reader-onnx | Native C++ JNI bridge reader jni.cpp linking against :lib-sherpa-onnx 's libonnxruntime.so OrtGetApiBase and static SentencePiece model.SRC / model.TGT . | | :lib-llama-android https://github.com/TinkeringDao/satori/blob/main/lib-llama-android | Native CMake build of llama.cpp compiling per-microarchitecture ARM64 libraries armv8.0 through armv9.2 with dotprod, i8mm, and FP16 acceleration loaded dynamically at runtime via /proc/cpuinfo detection. | - JDK 17 - Android SDK : Platform android-35 compile/target SDK 35, minSdk 29 , Build-Tools 34.0.0 - Android NDK : 29.0.13113456 and CMake 3.31.6+ for :lib-llama-android and :lib-reader-onnx - Git submodules : third party/llama.cpp and third party/sentencepiece must be initialized before building. Initialize native submodules llama.cpp and sentencepiece git submodule update --init --recursive Run unit tests across all modules 4,000+ unit tests ./gradlew testDebugUnitTest Run Android Lint checks ./gradlew lintDebug Build the debug APK ./gradlew assembleDebug The compiled debug APK is written to app/build/outputs/apk/debug/app-debug.apk . - GLOSSARY.md https://github.com/TinkeringDao/satori/blob/main/GLOSSARY.md — Domain, hardware, and on-device inference terminology across the codebase. - docs/GRILLE DESIGN.md https://github.com/TinkeringDao/satori/blob/main/docs/GRILLE DESIGN.md — Complete specification of the Grille visual system, color tokens, typography, and cover/inner screen layouts. - docs/QUICK KEY.md https://github.com/TinkeringDao/satori/blob/main/docs/QUICK KEY.md — Hardware power-key double-press com.motorola.mykey integration and lock-screen capture behavior. - docs/MODEL UX SPEC.md https://github.com/TinkeringDao/satori/blob/main/docs/MODEL UX SPEC.md — First-run onboarding, background download state machine, storage checks, and HomeStrip nudges. - docs/PRIVACY AND TERMS.md https://github.com/TinkeringDao/satori/blob/main/docs/PRIVACY AND TERMS.md — Privacy architecture and local-only data storage guarantees. - docs/THIRD PARTY NOTICES.md https://github.com/TinkeringDao/satori/blob/main/docs/THIRD PARTY NOTICES.md — Licenses and attributions for bundled libraries, fonts, GeoNames data, and downloadable models.