Execution layer
Sandboxed interpreter / executor. Runs only after intent validation, typically inside Docker on a companion host or supported on-device VM.
AMOS is a fully open-source floating command layer for Android, built on 01 (execution) and 02 (intent bridge). No subscriptions, no paid tiers — clone the repo, run the CLI, and self-host the backend.
AMOS = Autonomous Mobile Operating System — governed by your confirmation, never silent, never hijacking other apps.
npm install -g amos-cli
Optional PC Companion on Windows pairs with your self-hosted backend — no account fees.
AMOS is an open-source AI command layer, automation assistant, and wearable control surface — not a hosted SaaS you subscribe to.
Clone the repo, run amos-cli, and self-host the backend on your own hardware.
Sandboxed interpreter / executor. Runs only after intent validation, typically inside Docker on a companion host or supported on-device VM.
Open-source schemas, provider interface, safety gate, verifiers. The skeleton developers extend and ship with AMOS.
Floating command UI on Android, optional PC Companion on Windows, glasses and AR previews. You run the backend; BYOK for AI providers.
Same governance, three surfaces. AMOS keeps clarification on the phone; glasses stay input/output first; AR shares the same plan / risk / confirm contract.
Bubble, quick panel, expanded plan review, execution status — designed to stay lightweight over your workflow.
Open phone mockShort status lines, confirmations, hands-free wake paths. Execution stays on the handset / companion.
Open glasses HUDFloating intent cards with risk pills, awaiting your confirm. Same open stack, wider canvas.
Open AR canvasThe same intent JSON, the same SafetyGate, the same recycle controls — whether you tapped the bubble, said the wake phrase, shared from the browser, or spoke through your glasses.
AMOS is free and open source. Download the CLI, clone the repo, grab releases — no subscription required.
npm install -g amos-cli
amos init — create local .envamos backend start — API on :8000amos doctor — check your toolchainamos dev — full stack hintsOpen source
app/ in Android StudioOpen source
apps/pc-companion — .NET 8 WPFAMOS, 02 App Core, and 01 Executor are open for inspection, fork, and self-host. Contribute on GitHub — no commercial gate on features.
Developer path: CLI + clone + backend. Build Android from source or sideload. PC Companion pairs through your self-hosted API.
npm install -g amos-cli
git clone https://github.com/tensorverse/amos.git
cd amos
amos init
cd services/backend-api && pip install -r requirements.txt && cd ../..
amos backend start
Floating command layer + voice · routes to your paired PC through AMOS PC Companion.
Preview phone UI mock →Phone runs AMOS; glasses as display/audio/mic where supported.
Preview glasses HUD →Compatible devices run AMOS directly when supported.
Preview AR canvas →AMOS is open source — you host the backend and bring your own AI keys. No subscription or in-app purchase in this project.
The questions early testers ask first. Click any item to expand.
01 is the sandboxed executor — typically Docker on a companion host. 02 is the open intent bridge. AMOS is the open-source mobile / glasses / AR shell on top.
Yes — paste provider keys in settings. Keys stay on your device per in-app storage and privacy notices.
npm install -g amos-cli, clone the GitHub repo, run amos init and amos backend start, then build the Android app from app/ in Android Studio.
Android (the AMOS app) is the primary surface today. Glasses HUD and AR canvas previews are on this site; build from source for supported devices.
Usually no. The app exposes a Recycle control with four levels — Plan, Session, Executor, and All — that re-initialize specific layers. If a full reset is needed, uninstall and reinstall is supported and clears every setting and history; the only cost is re-pasting your API keys and re-granting overlay / foreground-service permissions. The Docker 01 executor lives on the companion host and is restarted independently with docker compose down && docker compose up -d.
No. AMOS is a user-authorized floating layer. The intent-only LLM contract (enforced by IntentContractValidator) blocks any executable payloads from the model — the LLM proposes intents, you confirm, and only then does 01 execute on a sandboxed runtime.
On device by default — DataStore for settings and API keys, Room for safety-decision history. The 01 executor runs in your own Docker on your own host. No telemetry is enabled at this milestone.
Open a GitHub issue or email support@tensorverse.ai.
Contribute connectors, automation packs, and docs on GitHub. No marketplace fees in the open-source tree.
Star on GitHubGitHub issues for bugs and features. Email for security or partnership questions.
support@tensorverse.ai