Jobs run on hosts near you, not in a distant data center. Less distance, faster responses.
Market-based pricing. Hosts set their own rates, you pay what compute actually costs. No cloud markup.
Every Cargo is signed and sandboxed. Hosts are identity-verified. No arbitrary code ever runs.
Run language models and image generators on community hardware. The coordinator finds the best Island for each job.
Email or GitHub. You get free beta credits to start.
Chat with an LLM, generate an image, or call the API from your code.
Responses arrive token-by-token from the nearest available Island.
Got a GPU sitting idle? Turn it into an Island and get paid when your machine does work for the network.
A lightweight Rust binary. Linux, Mac, or Windows.
Complete identity verification. Set your schedule and asking price.
The coordinator matches you with nearby consumers. You earn credits for each completed job.
The coordinator measures real RTT to every Island and places jobs on the lowest-latency match. Your request doesn't cross an ocean to reach a GPU.
Hosts set asking prices. Users set bid preferences. The coordinator clears a market rate for each Cargo type. No fixed tiers, no surprise bills.
Jobs default to regional hosts. For compliance, set explicit jurisdiction policies—EU-only, Swiss, or custom—and your data never leaves.
LLM responses stream token-by-token over WebSocket or SSE. Image generation reports progress. No polling, no waiting for a full response.
Identity verification via Stripe. Cargos are cosign-signed containers running under seccomp profiles. Hosts never see raw user data.
Every Island builds a public karma score from success rate, latency, and uptime. Higher karma means more jobs and better rates.
Models too large for one device run across multiple Islands in a pipeline. A 70B model on four laptops. Batch jobs fan out to fifty phones. Transparent to consumers.
Browse 133+ Cargos in the registry. Developers publish models and earn per execution. Reviews, ratings, and publisher tiers build trust.
iPhones and iPads become Islands via the iOS agent. On-device LLM inference with CoreML — no GPU workstation needed.
Submit up to 100 inputs in a single API call. Each dispatched to a different Island. Progress tracked in real time, results merged automatically.
Chain models into DAG pipelines. Steps run in parallel across Islands, data flows between them via templates, and results merge automatically.
Similar prompts served from cache instantly — free. Semantic matching catches near-duplicates. 30-60% cost reduction for repetitive workloads.
Deploy multiple model versions, split traffic, measure latency and accuracy per variant. Auto-promote the winner with statistical significance.
Train models across Islands without centralizing data. Each Island trains locally, shares only gradients. Your data never leaves your machines.
Encrypted end-to-end. Islands can't see your data, even while processing it. TEE hardware attestation or homomorphic encryption — your choice.
OpenAI-compatible chat endpoint. Switch your base URL and your existing code runs on community GPUs instead of a single cloud provider.
/v1/chat/completions# Works with any OpenAI-compatible client
from openai import OpenAI
client = OpenAI(
base_url="https://app.archipelag.io/api/v1",
api_key="your-key"
)
stream = client.chat.completions.create(
model="mistral-7b",
messages=[{"role": "user", "content": "Hello!"}],
stream=True
)
for chunk in stream:
print(chunk.choices[0].delta.content, end="")Turn any machine with a GPU into an Island. Set your own asking price and availability schedule. The coordinator matches you with nearby consumers who need compute.
No blockchain, no token. Just battle-tested infrastructure.
The coordinator handles placement, billing, and health. Elixir's actor model gives us per-job fault isolation and hot upgrades with zero downtime.
Job dispatch and output streaming over NATS. JetStream provides at-least-once delivery with consumer ack, so jobs don't get lost.
A lightweight binary that manages containers via Bollard, reports heartbeats, and streams output. Compiles to a single static binary.
We're opening Archipelag.io to everyone for three months. Come use AI, lend your GPU, and help us figure out what works.
AnnouncementWe built a distributed compute network that routes AI Cargos to nearby hardware instead of distant data centers. Here's the story of why, and what it actually does.
Create an account and run your first inference job on community hardware. Free beta credits, no credit card.
Create accountBuilding something that needs distributed inference? Work directly with our team on integration, custom Cargos, and priority access.
hey@archipelag.io