A decentralized intelligence ecosystem

The compute layer beneath every AI product.

HiveCompute routes AI workloads across decentralized networks — io.net and Together — with cryptographic settlement on Vitreus.

No vendor lock-in
No centralized control
No speculative token

Get an API key Explore the network

Plain English

What is HiveCompute, really?

For developers

One API. Every AI model.

No vendor lock-in. Write your code once and HiveCompute routes it to the best available GPU — automatically. The surface is a drop-in replacement for the OpenAI SDK; switch your base_url and you're live.

Start building →

For GPU owners

Your GPU earns while you sleep.

Connect your gaming PC or server to HiveCompute and get paid for every AI job it processes. The daemon installs in one command; jobs route to you when your card is the right fit.

Start earning →

For enterprises

Slash your AI costs by up to 97%.

Route your production workloads through HiveCompute and pay a fraction of what OpenAI, Anthropic, or Google charge. The market page shows live per-million-token rates side by side.

See the savings →

The Hive Ecosystem

HiveCompute is one layer of a three-layer decentralized intelligence platform. Each layer is independently useful. Together, they're what centralized AI infrastructure can't be.

HiveIQ

The intelligent routing layer. Agents, workflows, developer experience.

Live → hiveiq.xyz

HiveCompute

The decentralized compute layer. Providers, inference, fine-tuning, settlement.

You are here ▲

Vitreus

The trust and settlement layer. Identity, reputation, VTRS, on-chain finality.

1943

HiveCompute is the supply layer. HiveIQ routes to us. Vitreus settles what we do.

Watch a request route.

Every dot is a real request landing on /v1/chat/completions. Polled live from /v1/network/stats every five seconds — when the cumulative count moves, a pulse fires.

Awaiting first poll …

Real savings. Real time.

Every request you route through HiveCompute vs. going direct to OpenAI GPT-4o.

Total savings

—

Awaiting traffic

HiveCompute cost

—

OpenAI equivalent

—

Rates: io.net $0.0008/1k · Together $0.0002/1k · Groq $0.00059/1k vs OpenAI GPT-4o $0.005 / $0.015 per 1k input/output tokens

Your request. Anywhere.

HiveCompute routes to the nearest available provider — across decentralized networks and traditional cloud.

Built for every scale of provider.

From consumer hardware to enterprise-grade data centers. Four tiers, explicit criteria, honest trade-offs.

Flagship Tier 1

Enterprise

Eligibility: H100, H200, A100 80GB with NVIDIA CC
Verification: TEE (hardware-attested)
Uptime: 99.5% minimum
KYC: Business entity verification
Workloads: Regulated inference, private fine-tuning, enterprise LLM serving
Pricing: Premium

"Where enterprise compute actually lives."

TIER 2

Production

Eligibility: A100, H100, data-center hardware
Verification: Challenge-based with re-execution
Uptime: 98% minimum
KYC: None — staking and slashing
Workloads: Standard inference, fine-tuning, production deployments
Pricing: Standard

"Server-grade without the bureaucracy."

TIER 3

Community

Eligibility: RTX 4090, 3090, 6000 Ada
Verification: Challenge-based
Uptime: Best-effort
KYC: None
Workloads: Inference ≤13B, image gen, vision, TTS/STT
Pricing: Reduced

"Rideshare for compute, realized."

TIER 4

Experimental

Eligibility: Any CUDA-capable GPU
Verification: Challenge-based
Uptime: None required
KYC: None
Workloads: Bounty tasks, research, batch experiments
Pricing: Near-zero

"For when you just want to contribute."

Every tier uses the same API. Workloads are matched to the tier that fits — by cost, compliance, and performance requirement. You pick your tier. We pick your workload.

How it works.

Three steps to turn idle GPU hardware into VTRS earnings.

Register your hardware.

Tell us what GPU you have — anything from a home RTX 4090 to an H100 in a data center. Pick your tier based on what you're offering. Stake VTRS as trust collateral. No approvals, no gatekeepers.

Receive work automatically.

HiveCompute's routing engine matches incoming AI requests to your hardware based on capability, price, and reliability. You don't chase customers — the network brings jobs to you when your hardware fits.

Earn VTRS on every job.

Every completed job settles on-chain. Earnings flow to your wallet automatically, priced in USD and paid in VTRS at the current rate. Withdraw, hold, or stake more for higher-tier work.

Any CUDA-capable GPU qualifies. Server-grade hardware unlocks higher tiers and premium workloads.

Who this is for.

For developers

Drop-in AI compute, decentralized.

You're building with AI and your OpenAI bill is climbing. HiveCompute is a drop-in replacement — same /v1/chat/completions endpoint, same request format, same response shape. Route to models like Llama 3.3, DeepSeek, Qwen, and dozens more. Pay in VTRS at rates decided by a real marketplace, not a single vendor.

Get an API key

For providers

Earn from hardware you already own.

You have GPU hardware sitting idle — anything from a single RTX 4090 gaming rig to a data center of H100s. HiveCompute turns that idle time into VTRS earnings. Pick your tier, set your availability, accept the workloads you want. We handle routing, settlement, and everything in between.

Become a provider

Live · running on HiveCompute

Talk to the network.

The assistant on the right is served by HiveCompute itself — your message is routed through /v1/chat/completions to a real GPU provider, the same way any developer's request would be. Ask about the four provider tiers, the parachain plan, how integration works, anything.

Open the full chat

Ask about HiveCompute

Pick a starter or type your own.

The network is live.

Real numbers, updated in real time. No demo data.

Providers Online

Models Available

Requests Served

Total VTRS Settled

0.00 VTRS

Avg Latency p50

—

Phase 2.6

System operational. Polling /v1/network/stats.

Data pulled directly from the HiveCompute production database. No mock data, no fake activity, no paid signups. What you see is what's actually happening.

Developer API.

OpenAI-compatible. Bring your existing code.

curl https://hivecompute.xyz/v1/chat/completions \
  -H "Authorization: Bearer $HIVECOMPUTE_KEY" \
  -d '{
    "model": "meta-llama/Llama-3.3-70B-Instruct",
    "messages": [{"role": "user", "content": "Hello, Hive."}]
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://hivecompute.xyz/v1",
    api_key="hc_...",
)

resp = client.chat.completions.create(
    model="meta-llama/Llama-3.3-70B-Instruct",
    messages=[{"role": "user", "content": "Hello, Hive."}],
)
print(resp.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://hivecompute.xyz/v1",
  apiKey: "hc_...",
});

const resp = await client.chat.completions.create({
  model: "meta-llama/Llama-3.3-70B-Instruct",
  messages: [{ role: "user", content: "Hello, Hive." }],
});
console.log(resp.choices[0].message.content);

Example response text/event-stream

▍

Illustrative — this is a canned sample of the SSE output, not a live request.

Method	Endpoint	Status
GET	/health	Live
POST	/v1/chat/completions	Live
GET	/v1/network/stats	Live
GET	/v1/network/recent	Live
POST	/v1/embeddings	Live
GET	/v1/models	Live
POST	/v1/images/generations	Live
GET	/v1/providers	Live

Full reference at /docs. We ship endpoints in order of customer demand, not marketing convenience.