AINode is a free, open-source local AI platform that turns any NVIDIA GPU machine into a complete AI stack — browser chat UI, OpenAI-compatible REST API, and LoRA fine-tuning — with automatic multi-node clustering. It installs with one command and is licensed Apache 2.0.

How is AINode different from Ollama, LM Studio, or text-generation-webui?

AINode is designed for multi-GPU and multi-node clusters from the start. It runs vLLM as the inference engine, Ray for cross-node orchestration, and a patched NCCL for GB10 hardware, so one large model can be sharded across every GPU on your network. Ollama and LM Studio target single-machine use; AINode targets the DGX Spark, ASUS GX10, and any other Linux host with an NVIDIA GPU that you want to scale horizontally.

What hardware do I need to run AINode?

AINode runs on any Linux host with an NVIDIA GPU and CUDA 13 drivers. It is verified on NVIDIA DGX Spark and ASUS GX10 (GB10, 128 GB unified memory). For a distributed cluster, a low-latency RoCE or InfiniBand fabric (e.g., ConnectX-7 at 200 Gbps) is recommended; 10 GbE also works for smaller models.

Does AINode support distributed inference across multiple GPUs or nodes?

Yes. v0.4.0 ships real tensor-parallel distributed inference across multiple nodes, verified on two DGX Sparks with 244 GB aggregated VRAM. The head node auto-discovers peers over UDP on port 5679, bootstraps SSH, and uses Ray + NCCL (over RoCE where available) to shard one model across every GPU in the cluster.

How do I install AINode?

Run: curl -fsSL https://ainode.dev/install | bash. The installer checks prerequisites, pulls the ghcr.io/getainode/ainode container image, installs a systemd service, and starts the stack. The chat UI is at http://localhost:3000 and the OpenAI-compatible API at http://localhost:8000/v1.

Yes. AINode is 100% free and open source under the Apache 2.0 license. There are no tiers, no cloud dependencies, and no recurring costs. You own the hardware, you own the data.

Which models does AINode support?

AINode supports any model compatible with vLLM, including Meta Llama 3 and 3.1, Mistral and Mixtral, Qwen 2.5, Google Gemma 2, DeepSeek V3, Phi-3, and most open-weight models on Hugging Face. The model download UI lets you paste any Hugging Face repo ID.

Where do I pull the AINode container image?

AINode is distributed as a unified container image on two registries. The canonical pull is ghcr.io/getainode/ainode:latest (used by the installer). A Docker Hub mirror lives at argentaios/ainode:latest for discoverability.

Can I fine-tune models with AINode?

Yes. AINode ships a browser-based fine-tuning panel with LoRA and full-fine-tune templates, dataset upload, hyperparameter configuration, and live loss curves. Fine-tuning runs on the same GPU(s) used for inference — your hardware, your compute, zero per-token cost.

How do I update AINode?

Run ainode update on any installed host. That pulls the latest container image from GHCR and restarts the systemd service. Configuration, model weights, and fine-tune outputs live on the host, so upgrades never touch your data.

AINode is a free, open-source local AI platform (Apache 2.0) that turns any NVIDIA GPU machine into a complete AI stack: browser chat UI on port 3000, OpenAI-compatible REST API on port 8000, LoRA fine-tuning, and automatic multi-node clustering with tensor-parallel distributed inference over NCCL. v0.4.0 is verified on NVIDIA DGX Spark (GB10, 128 GB unified memory) with 244 GB aggregated VRAM across two nodes. Install with one command: curl -fsSL https://ainode.dev/install | bash.

v0.4.0 released — Apache 2.0

Turn any NVIDIA GPU into
your local AI platform.

Chat, API, and fine-tuning in your browser.
One command to start.

AINode 4-node cluster — 3 DGX Sparks + ASUS GX10, 487 GB aggregated VRAM, live topology with pulsing connections

Four GB10 nodes, 487 GB aggregated VRAM, automatic discovery.

Terminal

Get running in minutes

How It Works

Install

One command. No dependencies, no Docker, no fuss.

$curl -fsSL https://ainode.dev/install | bash

Start

Launch the platform and open your browser.

$ainode start

Use

Chat, call the API, or fine-tune models — all from the browser.

The whole platform, in one binary

Chat, API, downloads, fine-tuning, and cluster config — all shipped in the same container. No separate services to wire up.

AINode chat UI streaming a response from a locally hosted model — Chat in your browser
OpenAI-compatible streaming chat, served from your own hardware on port 3000.

AINode server API console with live request logs — Live API console
Drop-in OpenAI-compatible /v1 endpoints. Watch requests stream through in real time.

AINode model download manager with 50+ supported models — 50+ models, one click
Browse the curated catalog or paste any Hugging Face ID. Downloads run in the background.

AINode cluster configuration screen showing tensor parallel and minimum node settings — Configure the cluster
Set minimum nodes, pick tensor or pipeline parallel, and AINode auto-shards across every GPU it discovers.

AINode fine-tuning dashboard overview — Fine-tune, in the browser
LoRA and full fine-tunes with dataset upload, eval, and live loss curves — no notebooks required.

AINode fine-tuning templates gallery — Templates for common tasks
Instruction tuning, classification, summarization — pick a template and go.

Everything you need

Batteries included

Chat UI

Chat with AI models in your browser

OpenAI-compatible API

Drop-in replacement for OpenAI endpoints

Model Fine-tuning

Fine-tune models on your own data, from the browser

Multi-node Clustering

Automatic discovery and load balancing across nodes

50+ Models

Llama, Mistral, Gemma, Qwen, and more

Built on vLLM

Production-grade inference engine under the hood

The case for local

Why Local AI?

	Cloud AI (OpenAI, etc.)	AINode
Monthly cost	$100–$10,000+	$0
Data privacy	Their servers	Your hardware
Rate limits	Yes	No
Latency	200–2,000 ms	10–50 ms
Fine-tuning	Limited, expensive	Unlimited, free
Internet required	Yes	No

Monthly cost

$100–$10,000+

Data privacy

Their servers

Your hardware

Rate limits

Yes

Latency

200–2,000 ms

10–50 ms

Fine-tuning

Limited, expensive

Unlimited, free

Internet required

Yes

Compatible Hardware

Built for NVIDIA. Runs everywhere.

From a $2,999 DGX Spark to your existing Linux workstation — if it has an NVIDIA GPU, AINode runs on it.

NVIDIA DGX Spark

128GB unified memory

From $2,999

ASUS AI Server (GB10)

Grace Blackwell platform

Enterprise

Dell AI Factory

PowerEdge + NVIDIA

Enterprise

HP AI Workstations

Z-series with NVIDIA

Professional

Any Linux + NVIDIA GPU

Ubuntu, RHEL, Debian

Your hardware

Starting at $2,999Less than half the price of a Mac Studio

Cost Calculator

How much could you save?

Compare your monthly cloud AI spend against a one-time hardware investment.

Monthly cloud AI spend$500

$100$10,000

Annual cloud cost

$6,000

Hardware (one-time)

$3,499

Break-even

7 mo

3-year savings

$14,501

3-year cloud cost$18,000

AINode (one-time)$3,499

Built on the best of the community

AINode was born from the strong community foundation around DGX Spark. We build directly on the battle-tested patched NCCL and recipes from eugr/spark-vllm-docker — the de facto standard for reliable multi-node vLLM on GB10 hardware.

We then added the polished UX layer: beautiful cluster visualization, one-command install, browser-based training, and a clean OpenAI-compatible API. The result is a complete local AI platform that feels approachable — even if you just dropped a few thousand on one or two Sparks.

Huge thanks to eugr and the contributors making multi-node Spark setups practical.

Open source. Community driven.

AINode is free, transparent, and built in the open. No vendor lock‑in, no subscription traps.

Star on GitHubApache 2.0Free & Open Source

Testimonial coming soon

Install in 30 seconds

Pick your role. One command starts everything — Docker image, systemd service, and web UI on port 3000.

One GPU, standalone node

$ curl -fsSL https://ainode.dev/install | bash

Apache 2.0 · Powered by argentos.ai · Runs on any NVIDIA GPU · docs.ainode.dev

Sunbelt Computer Software

PL/B Language Development and Support

Turn any NVIDIA GPU into
your local AI platform.