Serve & Earn — zero install
Your phone is an AI miner now
This page turns any modern browser into an Animica inference worker — a phone on a shelf, a laptop on a desk, a gaming PC overnight. A small open model downloads once and runs entirely on your device (GPU via WebGPU on Chrome/Edge; CPU via WebAssembly everywhere else, iPhone included); the page claims real chat jobs from the network, answers them locally, and credits your wallet in ANM per job won. Nothing to install, no keys on this page — just an address to pay.
Serving counts toward the network's inference carve — 75 ANM per block (25% of the block reward): every block with ANY new inference gets a settlement anchor that moves the whole carve to providers, split pro-rata by earned weight; blocks without inference roll it to the treasury.
How it works
- The model runs on your device. WebLLM compiles a quantized open model (Qwen 2.5, ~1 GB download, cached by your browser) to WebGPU. Prompts and answers never leave your device except to deliver the finished answer to the job queue.
- Jobs come from the shared AICF queue. The same queue every
animica upnode serves — free-chat traffic from animica.dev and paid inference jobs. Your browser polls, claims, answers, submits. - Races keep quality honest. Jobs are replicated to several workers; the first good answer wins and is the one credited. A fast desktop GPU will usually beat your phone when both are online — and your phone still wins whenever it's the fastest (or only) worker awake.
- Teams pool token budgets. Long-form requests fan out across every machine online — one worker drafts the outline, others each write a section in parallel, and the network stitches one long answer. More devices serving means more sections, more depth, and more winners paid per question.
- Earnings are per-job IOUs. Each job won credits its full estimated cost to your address on the node's worker ledger (
aicf.workerEarnings). Settlement to on-chain ANM rides the AICF service carve as it settles network-wide.
Prefer a terminal? The Termux lane
Same queue, same ledger, no browser: a dependency-free package that drives llama.cpp natively. Survives in a tmux session with the screen off (termux-wake-lock), and works on any Linux box the same way.
pkg install python llama-cpp # Termux (Android)
pip install animica-serve
animica-serve --address anim1yourwallet--model qwen2.5-0.5bfor older / low-RAM phones ·--charge-onlypauses while unplugged (needs the Termux:API app).- Already running Ollama or a llama-server?
--openai-url http://127.0.0.1:11434/v1 --openai-model qwen2.5:1.5breuses it — no download. animica-serve earnings --address anim1…prints your ledger any time.
Requirements & battery
- Any modern browser. With WebGPU (Chrome/Edge on Android or desktop) the model runs on your GPU. Without it — iPhone Safari included — the page automatically switches to a CPU (WebAssembly) engine: slower, but it serves and earns the same way. Pick the 0.5B model on phones with less RAM.
- Plugged in; screen your choice. Sustained inference is real work: leave the phone charging (it auto-pauses when unplugged where the browser exposes battery state). "Keep screen awake" holds a wake-lock — the most reliable mode. Turn it off to let the screen sleep: on Android the background worker + keepalive tone keep serving; iPhone suspends sleeping tabs, so leave it on there.
- Background serving. The worker runs on a background thread, so on Android Chrome and desktop it keeps serving while you use other apps — leave "keep serving in background" on (it plays a near-silent tone so the browser doesn't freeze the tab). iOS suspends background tabs regardless: on iPhone, leave the tab open on the charger.