Nataris — P2P inference via Android phones, works as an OpenAI-compatible API

[Sharrmavishal]

Built this over three months with a two-person team. Jan users already understand the value of running AI locally — Nataris extends that idea into a network where anyone's Android phone can serve inference to developers via an API.

What Nataris is

A P2P inference marketplace. Android phones run open-weight models locally (Qwen 2.5 0.5B, Llama 3.2 1B) and serve API requests via a standard OpenAI-compatible endpoint. Phone owners earn per token. Developers get inference without managing any servers.

No prompt logging. No content filtering. No model training on your queries.

How it connects to Jan

Jan is about local-first AI. Nataris takes that further — your phone becomes part of a shared compute network when you're not using it. It's the same ethos: open-weight models, user-controlled, privacy-respecting.

For developers: if you're already using Jan for local inference but need an API for production, Nataris gives you that without going through a centralized provider.

Quick start

Sign up at nataris.ai — $5 free credits on signup, no card needed.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.nataris.ai/v1",
    api_key="YOUR_NATARIS_API_KEY"
)

response = client.chat.completions.create(
    model="nataris-fast",
    messages=[{"role": "user", "content": "Hello!"}],
    stream=True
)

nataris-fast = Qwen 2.5 0.5B (quick, ~5s)
nataris-balanced = Llama 3.2 1B (better quality, ~15-20s)

Want to be a provider?

Install the Android app and earn while your phone is idle. App on Google Play: https://play.google.com/store/apps/details?id=ai.nataris.app

Where we are

21 providers on the network, 2,775 inference jobs completed, 350K+ tokens processed. Just launched publicly.

Happy to answer questions.