Run any Python function on a decentralized GPU fleet

Animica Studio is a Modal-style serverless platform for the post-quantum chain. Write a function, add a decorator, and run it on Animica's GPU & CPU fleet — .remote(), .map(), cron, and web endpoints — metered and settled on-chain in ANM. The same rigs that mine and train ANM now run your code.

Open docs → View on GitHub $ pip install animica

app.py

import animica.studio as studio

app = studio.App("trainer")

@app.function(image=studio.Image.debian_slim().pip_install("torch"), gpu="A100", timeout=900)
def train(seed: int) -> float:
    import torch
    return float(torch.randn(1024, seed % 8 + 1).mean())

train.remote(42)          # one call on the fleet
train.map(range(128))     # fan-out across providers

How it works

From zero to running in three steps

No servers, no YAML, no Dockerfiles. Install the SDK, decorate a function, and call it.

Install

pip install animica

The complete client — Studio SDK + native CPU miner included by default. For every optional extra (Qt-wallet QR, the full distributed Studio client, all server deps): pip install "animica[all]".

Decorate

@app.function(gpu="A100")
def train(seed): ...

Run

train.remote(42)
animica studio run app.py::train

Capabilities

Everything you expect from serverless

@app.function

Decorate any Python function and run it remotely — no servers, no YAML, no Dockerfiles required.

.remote() & .map()

Call a function once on the fleet, or fan out across thousands of inputs with one line.

GPU on demand

Pin A100/H100-class GPUs per function. Pay per second of compute, settled in ANM.

Images in Python

Declare pip/apt deps inline with Image.debian_slim().pip_install(...). Deterministic, content-addressed.

Secrets & Volumes

Inject secrets into a function and mount persistent volumes — no plaintext in your job.

Cron & web endpoints

Schedule functions on a cron, or expose them as HTTP endpoints for your app and agents.

Decentralized fleet

The same rigs that mine and train ANM run your functions — useful work, not wasted hashes.

On-chain settlement

Every run is metered and paid in ANM on a live post-quantum L1. Verifiable in the explorer.

Positioning

Animica Studio vs. Modal

If you know Modal, you already know Studio — the SDK is intentionally familiar. The difference is where your functions run and how they're paid for.

Decorator-based Python SDK	Animica Studio	Modal
Serverless GPU functions	✓	✓
.remote() / .map() ergonomics	✓	✓
Cron + web endpoints	✓	✓
Pay-per-second billing	ANM (on-chain)	USD (card)
Runs on a decentralized fleet	✓ miners & GPU providers	✗ centralized cloud
Useful-work mining synergy	✓ same rig mines + serves	—
Post-quantum, verifiable settlement	✓	—
Open source	✓ Apache-2.0	partial

For AI agents

Discoverable and callable by autonomous agents

Studio ships an llms.txt manifest and a Model Context Protocol (MCP) server so AI agents can discover the platform and run or deploy compute on their own — paying from a wallet.

studio_runRun a function once on the fleet (or in a local sandbox).

studio_deployRegister a named, versioned app from source.

studio_listList the functions an app has deployed.

studio_estimateQuote the ANM cost of a run before paying.

Point your agent here

animica-mcp   # npm
animica_mcp   # PyPI

# manifest for LLMs / crawlers:
https://studio.animica.org/llms.txt

Read /llms.txt →

Get started

Ship your first function in a minute

Local mode runs in a sandbox on your machine with zero infrastructure. Remote mode escrows ANM and dispatches to the live fleet.

Open docs → GitHub $ pip install animica