OGAM is an on-device AI suite for Android, iOS, and macOS that runs chat, image generation, vision, speech, tool calling, and knowledge base features offline. It supports GGUF models, multiple backends (CPU, GPU, NPU), and local servers, with memory management controls and per-model RAM accounting.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Off Grid AI (OGAM) is an offline, on-device AI suite for mobile and Mac platforms. It provides chat, image generation, vision, voice transcription, tool calling, and knowledge-base capabilities, all designed to run without network connectivity and without sending data off device. The project is implemented in TypeScript and licensed under MIT.
How it works
OGAM offers on-device processing across multiple components: text generation with support for multiple LLMs, on-device image generation, vision AI, and Whisper-based speech-to-text. It includes a model manager that supports memory-aware loading policies (Lean, Balanced, Aggressive) and provides a Load Anyway option when loads are refused. It can connect to local or remote OpenAI-compatible servers (Ollama, LM Studio, LocalAI) and supports tool calling with built-in tools. Documents can be uploaded to a project knowledge base and retrieved locally via cosine similarity using an embedded MiniLM model. downloads are managed with a dedicated download manager. The app can detect device capabilities (GPU, NPU, CPU) and select an appropriate backend, with options overrideable in Settings.
Getting started
Install options include platform app stores and a GitHub releases download. Build from source steps are provided:
git clone https://github.com/off-grid-ai/OGAM.git
cd OGAM
npm install
# Android
cd android && ./gradlew clean && cd ..
npm run android
# iOS
cd ios && pod install && cd ..
npm run ios
Requires Node.js 20+, JDK 17 / Android SDK 36 (Android), Xcode 15+ (iOS).
Install instructions and build prerequisites are stated in the README as shown above.
Recent releases
Latest release: v0.0.103 Off Grid 0.0.103 (2026-07-15). This production release bundles recent work focusing on memory control and stability. Also includes v0.0.103-beta.4 (2026-07-15), featuring new model-loading modes (Lean, Balanced, Aggressive) and updated model settings for transcription and text-to-speech. Prior beta releases include v0.0.103-beta.3 (2026-07-15) with similar loading mode updates and visibility into loaded RAM, and v0.0.103-beta.2 (2026-07-09) introducing a toggle to turn thinking on or off for reasoning models.
Traction
2868 stars as of the provided metrics. 138 open issues. 273 forks.
Behind the repo
Not provided in the excerpt.
Caveats
MIT license is stated. No exclusions beyond what's listed in the README excerpt; age: repo created 2026-01-29, last_push 2026-08-03. Open issues listed as 138. No further license or open issue caveats provided in the supplied material.