Private AI for your whole team

A self-hosted platform for running private LLMs and image models for one person, a team or an organization. Role-based access, load balancing with failover, and nothing leaves your network.

Read the docs GitHub

What it does

Model management

Browse and download GGUF models straight from Hugging Face.

Streaming chat

Attachments, per-chat model switching and tunable generation settings.

Multi-user channels

Private conversations, shared channels and per-channel managers.

Knowledge base

Add your documents and get answers that cite their sources.

Tools via MCP

Connect MCP tool servers and let models call them in chat.

Live dashboards

Usage, telemetry, and CPU, RAM and GPU monitoring.

Load balancing

Spread requests across engines, with automatic failover.

Cluster management

Manage a Matricxon cluster on a live node map (BETA).

Secure by design

Hashed passwords, signed sessions, rate limits and roles.

Three ways to run it

The big choice is where the heavy model work runs: on the same node as pAIring, on a shared cluster, or on separate servers.

1. Single node

Browser ─► [ One node ]
           pAIring + engine
           + models

Everything runs on one node, and pAIring manages the engine for you.
One install, nothing else to set up.
Capacity is that node: its RAM and GPU hold every model.

Best for personal use, small teams and trying it out.

2. Matricxon cluster (BETA)

pAIring ─┬─► Node A ─┐
         ├─► Node B ─┼► shared
         └─► Node C ─┘  folders

Nodes share one models folder and one chat-cache folder.
A chat can move to any free node without re-reading its history.
Real load balancing, and the cluster keeps serving if a node fails.

Best for many users who need steady speed.

3. Separate servers

pAIring ─┬─► Server A
         ├─► Server B
         └─► Server C

Independent engine servers. Models and chat caches are not shared.
Requests are spread out and failover works, but chats tend to stick to one server.
Works with Ollama or Matricxon.

Best for cheap extra capacity and failover.

Compare the layouts in detail

Quick start

python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
cp .env.example .env
.venv/bin/python run.py

More options and configuration are in the documentation.

Also see Matricxon