Backend services temporarily down. Will be up soon. Join the waitlistSupport

Turn idle compute into income.

Access LLMs for less. Earn from spare CPU/GPU or call open-source models through one unified API. We're in beta — join the waitlist to get early access.

Join API Waitlist

Beta · Waitlist open

GPT-4.1OpenAIFrontier
o3OpenAIReasoning
Claude 4AnthropicFrontier
Claude Sonnet 4AnthropicFrontier
Gemini 2.5 ProGoogleFrontier
Gemma 3GoogleOpen
Llama 4MetaOpen
Llama 3.3 70BMetaOpen
DeepSeek R1DeepSeekReasoning
DeepSeek V3DeepSeekOpen
Qwen 3AlibabaOpen
Mistral LargeMistral AIFrontier
Mixtral 8x22BMistral AIOpen
GrokxAIFrontier
Command R+CohereEnterprise
Phi-4MicrosoftEfficient
GLM-4Zhipu AIOpen
Kimi K2Moonshot AIFrontier
GPT-4.1OpenAIFrontier
o3OpenAIReasoning
Claude 4AnthropicFrontier
Claude Sonnet 4AnthropicFrontier
Gemini 2.5 ProGoogleFrontier
Gemma 3GoogleOpen
Llama 4MetaOpen
Llama 3.3 70BMetaOpen
DeepSeek R1DeepSeekReasoning
DeepSeek V3DeepSeekOpen
Qwen 3AlibabaOpen
Mistral LargeMistral AIFrontier
Mixtral 8x22BMistral AIOpen
GrokxAIFrontier
Command R+CohereEnterprise
Phi-4MicrosoftEfficient
GLM-4Zhipu AIOpen
Kimi K2Moonshot AIFrontier

100+ models, one decentralized network

Access frontier and open-weight models through one unified API — routed across the decentralized Kielo network.

GPT ImageOpenAIText-to-Image
FLUX ProBlack Forest LabsText-to-Image
Ideogram 2IdeogramText-to-Image
Stable Diffusion 3.5Stability AIText-to-Image
Recraft V3RecraftText-to-Image
Imagen 3GoogleText-to-Image
Veo 2GoogleText-to-Video
Kling 1.6KuaishouText-to-Video
Hunyuan VideoTencentText-to-Video
Wan 2.1AlibabaText-to-Video
SeedanceByteDanceText-to-Video
LTX VideoLightricksText-to-Video
CogVideoXZhipu AIText-to-Video
MochiGenmoText-to-Video
GPT ImageOpenAIText-to-Image
FLUX ProBlack Forest LabsText-to-Image
Ideogram 2IdeogramText-to-Image
Stable Diffusion 3.5Stability AIText-to-Image
Recraft V3RecraftText-to-Image
Imagen 3GoogleText-to-Image
Veo 2GoogleText-to-Video
Kling 1.6KuaishouText-to-Video
Hunyuan VideoTencentText-to-Video
Wan 2.1AlibabaText-to-Video
SeedanceByteDanceText-to-Video
LTX VideoLightricksText-to-Video
CogVideoXZhipu AIText-to-Video
MochiGenmoText-to-Video
ElevenLabs TurboElevenLabsTTS
Cartesia SonicCartesiaTTS
PlayAI DialogPlayAITTS
Whisper LargeOpenAISTT
Gemini Live AudioGoogleSTT
Deepgram NovaDeepgramSTT
Jina Embeddings v3Jina AIEmbeddings
Voyage 3Voyage AIEmbeddings
Cohere Embed v4CohereEmbeddings
BGE M3BAAIEmbeddings
Cohere Rerank 4CohereReranker
GPT-4o VisionOpenAIVision
Claude VisionAnthropicVision
Gemini VisionGoogleVision
PaddleOCRBaiduOCR
GPT-4oOpenAIMultimodal
ElevenLabs TurboElevenLabsTTS
Cartesia SonicCartesiaTTS
PlayAI DialogPlayAITTS
Whisper LargeOpenAISTT
Gemini Live AudioGoogleSTT
Deepgram NovaDeepgramSTT
Jina Embeddings v3Jina AIEmbeddings
Voyage 3Voyage AIEmbeddings
Cohere Embed v4CohereEmbeddings
BGE M3BAAIEmbeddings
Cohere Rerank 4CohereReranker
GPT-4o VisionOpenAIVision
Claude VisionAnthropicVision
Gemini VisionGoogleVision
PaddleOCRBaiduOCR
GPT-4oOpenAIMultimodal

100+ models across LLMs, vision, video, audio, and AI services

How it works

With Kielo you can run AI on a decentralized network of real machines — or put your own hardware to work. Two parallel paths, one network.

Run open models

One API. Any open model.

Generate an API key and call Llama, Mistral, Qwen, DeepSeek, and 100+ more through a single OpenAI-compatible endpoint. Your existing SDK works — change one line.

main.py
# Point your OpenAI SDK at Kielofrom openai import OpenAI client = OpenAI(    base_url="https://api.kielo.in/v1",    api_key="kielo_...",) res = client.chat.completions.create(    model="meta-llama/llama-3.3-70b",    messages=[{"role": "user", ...}],)

Intelligent routing

Routed to real hardware, everywhere.

Kielo scores every node on latency, reputation, and availability, then matches your request to the best machine on the network — with automatic failover when a node drops.

route.json
# Every request is routed to live, verified hardwareroute: {  "node":       "node_7f2a · RTX 4090",  "region":     "ap-south-1",  "latency_ms": 89,  "reputation": 0.98,  "failover":   true} # Unhealthy node? Kielo reroutes automatically.

Earn from idle compute

Your machine works while you don't.

Install the desktop app, benchmark your hardware, and accept inference jobs when your CPU and GPU sit idle. Transparent payouts, thermal safeguards, full control.

kielo-node
# Install Kielo Desktop Node (macOS arm64)curl -fsSL https://kielo.in/install.sh | bash  hardware detected   Apple M3 Pro · 18-core GPU benchmark score     8,420 (rate ×1.4) go live             sandboxed · thermal-safe jobs completed   142current rate     $4.28/hrsession earnings $31.06 ▂▄▆█▅▇

Unit Economics · Kielo

Provider Earnings Calculator

Input your hardware specs and see your potential earnings. All rates verified against market data as of Aug 2026.

Hardware

80%
0%
(0% for now)

Costs

Estimated Net Earnings / Month

+$179

Net $0.31/hr while active · 576 active hours/mo

GrossPlatform feeAfter

Gross Revenue

$202

Platform Fee

-$0

Host Payout

$202

Electricity

-$22

Net $/HR

$0.31

Annual ROI

15%

Payback Period

8.4 months

Against $1,500 hardware cost

Break-Even Utilization

11.0%

Below this, electricity costs exceed revenue

Projected Annual Net

$224/year

Total Active Hours/Year

6,912

Scale on Kielo

Thousands of provider machines, one unified API. The network absorbs your traffic spikes — you never think about GPUs, queues, or capacity planning.

  • Meta
  • Google
  • OpenAI
  • Anthropic
  • DeepSeek
  • Mistral AI
  • Alibaba
  • Cohere

Built for production load

12.4K output tokens per second at peak, spread across thousands of independent nodes instead of one datacenter queue.

Latency keeps dropping

89ms median time-to-first-token. Requests route to the nearest healthy node, so p50 improves as the network grows.

Pay for what you use

Blended $/1M tokens vs. comparable closed APIs. Per-token rates or unlimited plans — methodology published before launch.

No single point of failure

Distributed by design. When a node drops mid-request, Kielo reroutes to the next best machine automatically.

Targets shown are design goals for launch, benchmarked on 70B-class open models. Full methodology and reproducible configs published in our docs.

Imagine what you can build and earn on Kielo

Get started

Download the desktop app to earn from idle compute, or join the developer waitlist for API access.

Beta

Get started

Install a provider node with one terminal command, or join the developer waitlist for API access.

Compute providers

CLI install · v0.1.0

Install Kielo Desktop Node on macOS Apple Silicon, sign in, benchmark your GPU, download a model, and go live on the network.

  • Auto hardware benchmark
  • Thermal safeguards
  • Pause anytime

CLI installer (macOS arm64)

curl -fsSL https://kielo.in/install.sh | bash

Paste this in Terminal. It downloads the node and installs it to /Applications.

How to use it

  1. 1Run the command in Terminal (Apple Silicon Mac)
  2. 2Open Kielo Desktop Node from Applications
  3. 3Sign in with your provider account in the browser
  4. 4Complete onboarding: hardware scan, benchmark, model download
  5. 5Run preflight and go live
Full CLI install guide
Download for Linuxsoon

We'll email installation tips, earnings guides, and network announcements. Unsubscribe anytime.

Developers

One API · 100+ models

Join the waitlist for early API access. One unified endpoint for multiple open-source LLMs.

  • OpenAI-compatible
  • Up to 60% lower cost
  • Automatic failover

Llama, DeepSeek, Mistral, Qwen & more

API quickstart docs

We send occasional updates about your waitlist status and API access timeline — typically no more than one email per week. No spam. Unsubscribe anytime.

FAQ

Real objections, direct answers

Everything providers and developers ask before joining the network.

Kielo is a decentralized AI inference network. Compute providers contribute idle CPU/GPU capacity and earn money. Developers access multiple open-source LLMs through one unified API at lower cost than traditional closed providers.

Providers install the desktop app and join the network with benchmarked hardware. Developers send API requests. Kielo intelligently routes each request to an available provider, returns the response, and handles billing and payouts on both sides.

No. Kielo routes LLM inference requests — the same workloads developers send to APIs like OpenAI. Providers earn by running AI inference jobs, not by mining cryptocurrency.

Minimum and recommended CPU/GPU specs are being confirmed with engineering. The desktop app auto-detects your hardware and runs a benchmark to determine eligibility and earning rates. Details published before provider onboarding opens.

Earnings depend on your hardware benchmark score, availability, and network demand. The payout formula — percentage of inference revenue with benchmark multipliers — is published before launch. We will not publish illustrative earnings estimates without real data.

Providers earn a revenue share on completed jobs minus a disclosed platform fee. Payout schedule, minimum thresholds, and withdrawal process are defined in the provider agreement — published before you register hardware.

Inference workloads are comparable to other compute tasks. The app includes thermal safeguards and you control availability windows. Electricity costs are yours — the published payout formula lets you evaluate whether participation makes sense for your setup.

Yes. You decide when your hardware is available for jobs. Pause or stop at any time from the desktop app — your machine only participates when you allow it.

Inference jobs run in sandboxed environments. Your personal files and unrelated processes are not exposed to network workloads. Local data stays on your machine.

We are finalizing the launch catalog of open-weight models (Llama, Mistral, Qwen, and others). The confirmed list will be published before your API batch opens.

One OpenAI-compatible endpoint gives you access to multiple open-source models. Generate an API key, choose a model, and send requests with your preferred SDK or HTTP client.

In addition to pay-as-you-go per-token pricing, Kielo offers unlimited usage plans at hourly, weekly, and other subscription tiers. Exact tiers and fair-use policies are published before launch.

Kielo targets up to 60% lower cost vs. comparable closed APIs on open-source models. The comparison methodology — which models, providers, and usage tiers — is published before launch. We do not present illustrative savings as final fact.

Any language that can make HTTP requests works. Kielo exposes an OpenAI-compatible API — use the OpenAI SDK with a different base URL.

Kielo monitors provider health continuously. If a node fails, the request is rerouted to another available provider where possible. Failover behavior is documented in the API reference before launch.

Enterprise SLA and self-hosted/on-prem deployments are on the roadmap for 2027. They are not committed for initial launch. Join the waitlist and note your requirements — we prioritize roadmap based on demand.

Still have questions?

We answer every email — usually within a day.

Talk to us