Contributed inference network

One API. A network of ordinary computers.

EasyAI routes OpenAI-compatible chat requests to spare GPUs on ordinary PCs. Contributors set the limits; the scheduler does the rest.

Windows 10/11 · invite-only · LFM3 live now.

Network

Status unavailable

  1. Client unknown
  2. Gateway unknown
  3. Scheduler unknown
  4. Worker unknown
Workers online
— of —
Models live
—
p50 first token
—

Workers online

—

active in the last 60s

Models live

—

serving at least one worker

Jobs 24h

—

user jobs, last 24 hours

p50 first token

—

median, last 24 hours

Call the API, or share spare capacity.

Both routes meet at the scheduler. One asks for inference; the other provides it.

Share spare capacity.

The worker app runs on your PC and accepts jobs only inside the limits you set.

  • Share GPU time when the machine is idle.
  • Set a VRAM budget, model-cache quota, idle-only wait, and active hours; CPU, power, thermal, and network ceilings are planned.
  • Pause, stop, or uninstall at any time.
Check your PC

Call the API.

Send an OpenAI-compatible request and stream the answer from a live model.

  • Chat completions with streaming responses.
  • LFM3-1.7B is live; Qwen3 models are planned.
  • Access is invite-only while the network grows.
Request API access

Three steps to contribute.

The worker app is a Windows service with a tray UI. You keep control of the limits; the scheduler handles placement.

  1. 01

    Check

    Run the local check. It measures your hardware and estimates a tier. Nothing is uploaded until you request an invite.

  2. 02

    Install

    After an invite, install the worker app and confirm the model and backend it should serve.

  3. 03

    Set limits

    Choose a VRAM budget, model-cache quota, idle-only wait, and active hours. Jobs outside them are refused on your machine; CPU, power, thermal, and network ceilings are planned.

One model live. Four planned.

LFM3-1.7B runs today. Qwen3 models are in planning; they are not available to call or host yet.

LFM3-1.7B

Catalog live
Params
1.7B
Min VRAM
~4 GB or CPU
Min RAM
8 GB
Backend
Ollama
Fits
CPU / 4–7 GB GPU

Planned

Qwen3-8B

Planned
Params
8B
Min VRAM
8 GB
Min RAM
16 GB
Backend
Ollama
Fits
Standard GPU

Qwen3-14B

Planned
Params
14B
Min VRAM
12 GB
Min RAM
24 GB
Backend
Ollama
Fits
Large GPU

Qwen3-30B-A3B

Planned
Params
30B-A3B
Min VRAM
20 GB
Min RAM
32 GB
Backend
Ollama
Fits
XL GPU

Qwen3-32B

Planned
Params
32B
Min VRAM
24 GB
Min RAM
32 GB
Backend
Ollama
Fits
XL+ GPU

Specs are estimates from 4-bit footprints plus ~15% headroom. The signed manifest is authoritative. Minimum specs are per worker, not per request.

Limits and failure handling are part of the design.

The worker is a guest on a contributor's PC. The scheduler plans around limits, heartbeats, and clean failure.

Limits are enforced locally.

The enable switch, VRAM budget, model-cache quota, idle-only wait, and active hours are checked on the worker machine. A job outside those limits is refused; CPU, power, thermal, and network ceilings are planned.

No prompt logging by default.

EasyAI does not log prompt text by default. The assigned worker sees the prompt and intermediate inference data to run the job.

Failed workers replan.

If a worker drops, the scheduler retries inside a bounded plan — at most two attempts — then fails the request cleanly.

POST /v1/chat/completions

Chat completions, OpenAI-compatible.

Send messages to a live model and stream the response. The request shape matches the OpenAI chat completions API; the model key is lfm3.

Request access
curl https://<your-domain>/v1/chat/completions \
  -H "Authorization: Bearer $EASYAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "lfm3",
    "messages": [
      { "role": "user", "content": "Explain leases in one paragraph." }
    ],
    "stream": true
  }'

Access is invite-only.

Tell us whether you want to call the API, contribute a PC, or both. We review requests manually and email invites when capacity opens.

Waitlist

Request an invite

Tell us whether you want to call the API, contribute a PC, or both. We review requests manually.

We use this address only to reply about access. Read the privacy notice.