Sign in

Use your email or service name to access the operator dashboard.

No account yet?

Register for Beta

Create a service account. You will receive a verification email, then your request will be reviewed before you can sign in.

3–64 characters: letters, numbers, hyphens, underscores.

At least 8 characters.

Already registered?

Verify your email

Verifying your email address…

Beta — experimental

The open source decentralized inference mesh

The next generation infrastructure for AI agents

A protocol for stitching together a global network of GPUs, turning idle compute into shared intelligence.

Up and running in minutes Ollama, vLLM & OpenAI-compatible Encrypted peer-to-peer inference Integrates into your current workflow
Soon

Windows

Soon

macOS

Soon

Linux

Docker

Docker Hub

mtrxAI is in public Beta. Features may change, break, or be discontinued without notice. Read Terms & Privacy.

Why we exist

Intelligence at the edge, shared by choice

"What if intelligence didn't live in one place, but in every place?"

For years, AI meant sending your questions to distant datacenters — convenient, but invisible. Who sees your data? Who sets the price? Who decides what's available?

Today, powerful models run on laptops and gaming PCs. A capable model fits on a desktop; smaller ones fit in your pocket. The unit of intelligence is coming home — and mtrxAI is how we connect those homes into something bigger.

Think of it as a neighborhood of minds: open your favorite app, ask a question, and never need to know whether the answer came from your GPU, a teammate's machine, or someone nearby with spare capacity. Same experience — you choose who you trust.

You join communities of shared compute on your terms — what you share, who you connect to, and when you pause. No monopoly. No black box. Just people helping people think faster.

What you get

Power when you need it. Privacy when you want it.

Browse models others are already running, attach your own, or ask someone nearby to warm one up. mtrxAI sits quietly between your apps and the mesh — you won't notice it's there until you need it.

Use or offer models

Find what's running on your network or across the mesh. Run locally when you can, borrow a heavier model from a friend when you can't.

Your machine first

If the model is on your GPU, it stays there. Fastest path, zero detours — your setup always wins when it can.

Finds the nearest help

Need a remote model? mtrxAI picks the closest available GPU — same room, same city, then the next best option.

Groups that fit you

Clusters — regional communities or private teams, smart ranking, fair credits. Swarms — token-based P2P circles with no central billing. Use either — or both.

Fits your workflow

Works with Cursor, VS Code agents, Open WebUI, and other tools you already use — no app changes required.

Fair exchange

When you borrow someone else's GPU, credits keep things balanced. Both sides verify usage before anything moves. Same team? Free by design. Beta tokens have no cash value.

Any backend, one door

Attach Ollama, vLLM, LM Studio, or your own API. Multiple servers, one local proxy — every app points at the same address.

Warm up what you need

Nobody running that 70B model? Ask the mesh to pull and load it on a GPU with enough VRAM — on the machine closest to you.

Share on your terms

Set hours when you're open for jobs. Thermal guard pauses sharing if your GPU runs hot. One switch stops everything.

The platform, at a glance

Mesh & routing

  • Local-first — your GPU wins when it can
  • Smart match: same network, nearest geo, lowest load
  • Clusters + swarms — run one or both
  • Public regions: Europe, Americas, Asia & more

Control & economy

  • Approve connections — or auto-approve your circle
  • Team dashboard: peers, credits, transactions
  • Per-model credit rates, fully transparent
  • Private groups with invite-only access

Built for builders

  • OpenAI & Ollama APIs, embeddings included
  • Coding agents: Responses API, tool calls, streaming
  • Searchable model catalog with VRAM estimates
  • Windows, macOS, Linux & Docker

Four steps to get started

Step 1

Install

Pull the Docker images and walk through setup. Save your ID — you'll use it to link other devices.

Step 2

Join a group

Pick a public regional community, join a private team with a shared invite, or enter a swarm token for fully decentralized sharing.

Step 3

Connect your models

Scan for Ollama or OpenAI-compatible servers on your machine. Attach one or many from the dashboard — mtrxAI merges them into a single catalog.

Step 4

Use — then share, if you want

Point your apps at mtrxAI. Browse what others offer. Share your GPU only when you're ready — approve each request, or turn on auto-approve.

Built for people like you

Developers & builders

One connection for your coding agents. Access models teammates host without copying huge files around.

Creators & designers

Borrow a heavier model on a friend's GPU for a big render or long session, then switch back to your own setup.

Gamers & PC enthusiasts

Your GPU sits idle between sessions. Share capacity in a private group and earn internal credits (Beta tokens, no cash value) for when you need someone else's power.

Teams & communities

Create a private group for your studio or friend circle. Only people you invite can see your models and connect.

Trust, by design

You choose your circle Public regional groups or private teams. People outside your group can't discover you.
Sharing is always opt-in Pick which models are visible. Accept or decline incoming requests. Schedule hours, pause, or stop cold.
Encrypted peer connections When E2EE is enabled, prompts travel peer-to-peer, encrypted. The lobby handles discovery; inference traffic is routed peer-to-peer. See Terms §2.
Verified identity Every peer carries a cryptographic device key. Production hosts can require attested client builds only.
Transparent credit accounting Credits settle only when both peers agree on usage. Mismatches are rejected — no silent charges. Beta tokens have no monetary value.
Block & ban Cut off anyone locally, instantly. Report abuse for Operator review.

Beta notice: This is experimental software. Do not send personal, confidential, or regulated data. Use is governed by our Terms & Privacy.

How it works

A mesh that thinks with you

You ask a question in your app. mtrxAI handles the rest — locally when it can, across the mesh when it must. You see a model name and an answer. The routing happens invisibly.

  1. You askYour app sends a question through mtrxAI, like always.
  2. Home firstIf the model is on your machine, the answer comes from your GPU. Done.
  3. Find a neighborModel not available? mtrxAI looks across your group for someone who has it running.
  4. Pick the closestIt chooses the nearest available GPU — less waiting, less lag.
  5. Direct connectionYour question and the answer travel directly between you and that person, encrypted.
  6. Keep it fairWhen you borrow compute, credits balance the exchange. Same team? Usually free.

Two ways to join

Cluster Swarm
Best for Regional communities and teams who want discovery, ranking, and fair credit tracking Small groups who want fully peer-to-peer sharing with no central billing
How you join Pick a public region or accept a private team invite Enter a shared token from someone in your group
Credits Tracked transparently when you borrow across teams Not tracked — honor system among peers
Connection Lobby signaling + direct WebRTC data channels libp2p gossip — fully decentralized
Mix & match Run clusters and swarms at the same time — regional discovery plus private P2P circles

What stays private

  • Your conversations — Inference goes peer-to-peer over encrypted channels. The coordination server handles discovery, not your prompts.
  • Your circle — You only see peers in groups you've joined. Private teams need an invite or shared secret.
  • Your keys — Room encryption derived from your group secret. Per-request session keys — nothing reusable across calls.
  • Your machine — Optional local proxy password. Bind to localhost by default. Your API keys stay in an encrypted local vault.
  • Your safety — Block locally, report globally. Thermal guard stops sharing before your hardware pays the price.

Where we're headed

Today — A living mesh

Multi-backend proxy, real-time discovery, E2EE inference, smart routing, team dashboards, and a credit ledger both sides must confirm.

Next — A fair compute economy

Reputation scores, dynamic pricing, confidential GPU attestation — experimental credit mechanics today; future features may change.

Always — Privacy and control

Local-first by default. You share what you want, when you want, with whom you want. Sovereignty isn't a feature — it's the foundation.

Try the mesh. Share when you're ready.

Install mtrxAI, join a community in your region, and browse what's already running. Use the network for a week before sharing anything — no commitment, full control.

This is experimental Beta software. Use at your own risk. See our Terms & Privacy before participating.

View on Docker Hub