Machine-Level Expansion

A single machine runs AI sessions. Multiple machines form a network of AI sessions. connector.json treats machines as deployment targets — same spec, any scale.

One Machine, One Daemon

Machine A:
  niia daemon (launchd, always running)
    ├── PTY Session 1: Claude
    ├── PTY Session 2: Codex
    └── PTY Session 3: Gemini

The daemon manages sessions. Headless server bridges them to the network. This is the atomic unit.

Two Machines, One Pipeline

Machine A (laptop):                Machine B (server):
  niia daemon                        niia daemon
    ├── Session: Claude                ├── Session: Codex
    └── Session: Haiku                 └── Session: ollama/llama3

  ←───── gateway relay (WSS) ─────→

{
  "machines": {
    "laptop": "MY-LAPTOP.local",
    "server": "BUILD-SERVER.local"
  },
  "pipeline": {
    "phases": [
      { "machine": "laptop", "model": "claude", "prompt": "Design the feature." },
      { "machine": "server", "model": "codex",  "prompt": "Implement it." },
      { "machine": "laptop", "model": "haiku",  "prompt": "Write tests." },
      { "machine": "server", "model": "ollama/llama3", "prompt": "Run on production data." }
    ]
  }
}

Same connector.json. Phases execute on different machines. The pipeline doesn’t know or care about machine boundaries.

N Machines, Full Mesh

Machine A ──── Machine B
    │    \    /    │
    │     \  /     │
    │      \/      │
    │      /\      │
    │     /  \     │
    │    /    \    │
Machine C ──── Machine D

Every machine talks to every machine. Every AI session on any machine can communicate with any AI session on any other machine.

{
  "machines": {
    "seoul":  "SEOUL.local",
    "tokyo":  "TOKYO.local",
    "sf":     "SF.local",
    "london": "LONDON.local"
  },
  "agents": [
    { "id": "kr-dev",  "machine": "seoul",  "model": "claude" },
    { "id": "jp-test", "machine": "tokyo",  "model": "gemini" },
    { "id": "us-ops",  "machine": "sf",     "model": "codex" },
    { "id": "uk-sec",  "machine": "london", "model": "ollama/llama3" }
  ],
  "communication": {
    "topology": "mesh"
  }
}

4 countries. 4 machines. 4 different AI models. Full mesh communication. One connector.json.

What Each Machine Provides

Machines aren’t interchangeable. Each has unique resources.

Machine	Has	Best for
Laptop	Fast network, developer context	Coordination, light tasks
GPU server	A100/H100, local models	Heavy inference, fine-tuned models
Data server	Production database, logs	Data analysis (data doesn’t move)
CI server	Build tools, test infrastructure	Testing, deployment
Air-gapped	Sensitive data, no internet	Regulated data processing

connector.json places AI where the resources are:

{
  "phases": [
    { "machine": "laptop",     "model": "haiku",          "prompt": "Coordinate." },
    { "machine": "gpu-server", "model": "ollama/llama-70b","prompt": "Heavy analysis." },
    { "machine": "data-server","model": "ollama/llama3",   "prompt": "Query prod data." },
    { "machine": "ci-server",  "model": "codex",          "prompt": "Run full test suite." }
  ]
}

Machine Lifecycle

Machines come and go. The daemon handles it.

# See what's online
niia remote status list
  SEOUL.local     [online]  3 sessions
  TOKYO.local     [online]  1 session
  SF.local        [offline]
  LONDON.local    [online]  2 sessions

# Start headless on a machine
niia remote start SF.local

# Upgrade niia on a remote machine
niia remote upgrade TOKYO.local

# A machine goes offline mid-pipeline
#   → failover to another machine with same capability
#   → or pause and resume when machine comes back

Machine as Deployment Target

In traditional infrastructure:

Kubernetes: "run this container on a node with 4 GPUs"

In connector.json:

connector.json: "run this AI on a machine with local LLM and production data access"

The machine is a deployment target. connector.json is the manifest. niia daemon is the runtime.

Kubernetes                    connector.json
─────────────                 ──────────────
Node                          Machine
Pod                           PTY Session
Container                     AI CLI (claude, codex, gemini)
YAML manifest                 connector.json
kubelet                       niia daemon
kubectl                       niia remote
Service mesh                  gateway relay + P2P

Same pattern. Different domain. Kubernetes orchestrates containers. connector.json orchestrates AI.

Get Started

Articles

So Anthropic

User Guide

Philosophy

Applied Philosophy

Architecture

Gestalt Color

Transparency

Terminal Core

Features

VariableMD

Research

CodexMono

OpenCLIs

vpncli

MonoSurf

AI-CLIs

Machine-Level Expansion

Machine-Level Expansion

One Machine, One Daemon

Two Machines, One Pipeline

N Machines, Full Mesh

What Each Machine Provides

Machine Lifecycle

Machine as Deployment Target

​Machine-Level Expansion

​One Machine, One Daemon

​Two Machines, One Pipeline

​N Machines, Full Mesh

​What Each Machine Provides

​Machine Lifecycle

​Machine as Deployment Target

Machine-Level Expansion

One Machine, One Daemon

Two Machines, One Pipeline

N Machines, Full Mesh

What Each Machine Provides

Machine Lifecycle

Machine as Deployment Target