AI Fusio
AiFusio MC
the agent console we build for our clients — now free for you

The platform behind AI Fusio's client work — now free for you

AI Fusio integrates AI into mid-market businesses every day. The implementations we ship aren't ChatGPT skins or no-code workflows — they're real production platforms with memory, multi-agent orchestration, audit trails, and cross-system integration. AiFusio MC is what we built for ourselves to do that work, and what we customize for our clients.

We open-sourced it for two reasons. First, it's the most honest demo of what AI Fusio actually delivers — see the engineering before you ever talk to us. Second, if you run a small team and want to start experimenting with agentic AI on your own data, you shouldn't have to pay $200/seat/month for tools that lock your data into someone else's cloud. The single-user version of this is a gift; the version customized to your company's systems, processes, and data — that's where we come in.

What it is

AiFusio MC (codename Fusio) is a local web app you run on your own machine. It orchestrates Claude, Codex, and any LLM via the MCP fabric with a curated skill library, durable SQLite memory, multi-agent missions, and an optional cross-machine bridge. What every commercial agent tool would ship if they didn't have multi-tenant concerns to worry about.

Runs on Linux, Windows, and macOS. Same app in the browser, same app as a mobile PWA, same app over SSH in a terminal. Your data lives at data/ — chats, memory, missions — all yours, all local, all backupable.

What's inside

Each of these is built into the app — no plugins, no upgrades, no Pro tier.

What it looks like

The Fusio design system: ink-dark background, signal-red accents, monospace and Space Grotesk typography, medallion-eyebrow-display headers, restrained palette.

Chat panel — sessions + project pill + model chips + tools menu

Luke's Chat — multi-agent missions with phase-by-phase live status

Terminal-look fullscreen mode

Tap the Full pill in any chat to enter terminal mode — monospace, no bubbles, no chrome, just text flowing. Same chat data, different surface. Designed for mobile where every pixel counts.

Install in 1 command (or 4, your call)

Two install paths. Both end at the same working install on localhost:3001.

npx @aifusiomc/create-fusio

— or, the manual way —

git clone https://github.com/reinvention18/fusio.git
cd fusio

npm install

npm run dev

# Just open http://localhost:3001 in your browser on any OS

What's actually running

The stack is intentional and locked in — one technology per concern, no microservice debugging, no orchestration overhead. Boots in under 5 seconds, survives weeks of uptime.

Ports it uses

Ports overridable via env vars (PORT, MC_HTTPS_PORT, MC_NEXT_PORT).

Where your data lives

data/
├── chats/<id>.json              # default-namespace chats
├── seo-chats/<id>.json          # SEO namespace
├── lukes-chats/<id>.json        # missions namespace
├── missions/<id>/
│   ├── state.json               # mission state
│   ├── events.jsonl             # event log
│   └── checkpoints/             # per-milestone snapshots
├── pending/
│   ├── <requestId>.json         # in-flight chat turn buffers
│   └── remote-chat/             # in-flight peer call state
├── memory.db (+ .wal, .shm)     # SQLite memory
├── notepads/<id>.json
├── uploads/
├── edit-log.jsonl               # cross-machine edit history
├── key-facts.json               # auto-injected facts
└── .secrets.json (chmod 600)    # encrypted credentials

All of it is just files. Back up data/ and you have a full restorable snapshot.

Tailscale — the recommended way to connect machines

You don't need Tailscale to run AiFusio MC, but if you want the multi-machine bridge or to install the mobile PWA, it's the simplest path. Tailscale is a free WireGuard mesh that gives every device a stable private IP and TLS certs, reachable from anywhere without port forwarding.

Why Tailscale specifically

• Stable private IPs that don't change when you move networks
• Free 90-day TLS certs via tailscale cert — solves the Chrome "PWA install needs HTTPS" problem in 30 seconds
• Encrypted end-to-end (WireGuard)
• No port forwarding, no dyndns, no VPN, no NAT punching
• Free for personal use, up to 100 devices

$ tailscale status
100.93.146.97   my-server   you@   linux   -
100.88.125.30   my-pc       you@   windows -

sudo tailscale cert my-server.tail123abc.ts.net
mv my-server.tail123abc.ts.net.crt certs/tls.crt
mv my-server.tail123abc.ts.net.key certs/tls.key
chmod 600 certs/tls.key

# In .env.local:
MC_ALLOWED_DEV_ORIGINS=https://my-server.tail123abc.ts.net:3443

pm2 start ecosystem.config.js
# Boots both 'mission-control' (Next on 3001) and 'mc-https' (proxy on 3443)
# Watch the boot log: "HTTPS proxy running on https://<your-host>:3443"

// On machine A (server):
{
  "myToken": "<random-base64>",
  "myLabel": "Server",
  "myUrl": "http://100.93.146.97:3001",
  "hosts": [
    { "id": "pc", "label": "Workstation",
      "url": "http://100.88.125.30:3001",
      "token": "<token-pc-issued>" }
  ]
}
// On machine B (PC): mirror it with reciprocal entries
{
  "myToken": "<token-pc-issued>",
  ...
  "hosts": [{ "id": "server", ..., "token": "<machine-A-myToken>" }]
}

Cert renewal (every 90 days)

sudo tailscale cert <your-tailscale-hostname>
# Replace certs/tls.crt and certs/tls.key, then:
pm2 restart mc-https

Or automate with cron: 0 2 1 * * sudo tailscale cert ... && pm2 restart mc-https renews on the 1st of every month at 2 AM.

Don't want Tailscale? Other options

• Single-machine local-only: skip Tailscale entirely. AiFusio MC runs fine on localhost:3001. No peer bridge or mobile PWA, but everything else works.
• Cloudflare Tunnel: free, similar mesh. Use a Cloudflare cert + MC_CERT_PATH / MC_KEY_PATH env vars.
• mkcert + LAN: home network only. Generate a self-signed cert with mkcert, install the CA on each device, point env vars at the mkcert output.
• Let's Encrypt: only if MC is on a real public domain — generally not recommended for a single-user tool.

Complete setup walkthrough

If the "4 commands" version (above) didn't get you fully set up, this is the full operator-grade walkthrough per OS.

System requirements

• Node.js 22+ (LTS recommended)
• ~4 GB RAM idle, 8 GB+ for active use (heap config raises Node to 8 GB)
• ~2 GB disk for node_modules + chat data growing over time
• Anthropic API key (pay-as-you-go; ~$5 funds weeks of light use)
• Optional: OpenAI API key for cross-provider scrutiny in missions
• Optional: Tailscale for multi-machine + mobile PWA

Linux step-by-step

# Ubuntu / Debian
curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt-get install -y nodejs git
sudo npm install -g pm2

# Fedora / RHEL
sudo dnf install -y nodejs git
sudo npm install -g pm2

cd ~
git clone https://github.com/reinvention18/fusio.git
cd fusio
npm install   # ~2-5 min, pulls Playwright + transformers

cp .env.example .env.local
# Edit .env.local — add ANTHROPIC_API_KEY=sk-ant-...

pm2 start ecosystem.config.js
pm2 save              # persist process list
pm2 startup           # generate systemd unit so MC survives reboot
# Follow the printed sudo command to enable

xdg-open http://localhost:3001
# Drop API keys into Settings → Credentials
# Add a workspace via the project pill in any chat header
# Send your first message

Windows step-by-step

npm install -g pm2 pm2-windows-startup
pm2-startup install

cd C:\
git clone https://github.com/reinvention18/fusio.git
cd fusio
npm install

copy .env.example .env.local
# Edit .env.local — add ANTHROPIC_API_KEY
pm2 start ecosystem.config.js
pm2 save

macOS step-by-step

brew install node@22 git
npm install -g pm2
cd ~
git clone https://github.com/reinvention18/fusio.git
cd fusio
npm install
cp .env.example .env.local
# edit .env.local, then:
pm2 start ecosystem.config.js

Verify it's running

pm2 status                                  # 'mission-control' should be online
curl http://localhost:3001/api/git/status   # should return JSON
open http://localhost:3001                  # macOS — open the UI

If something doesn't work — the fusio-debug Claude Code skill

The repo ships with a bundled Claude Code skill at .claude/skills/fusio-debug/SKILL.md that's specifically built to diagnose and fix AiFusio MC issues. When Claude Code runs in your cloned repo, the skill auto-loads on prompts like "fix mc", "mc is broken", "chat is hung", "mission stuck", etc.

The skill covers 13 known failure modes with diagnostic commands + fix sequences:

• Wedged PM2 (online but unreachable)
• White screen on load (V2 minifier TDZ pattern)
• Chat hung / stop button unresponsive
• Missions stuck
• Memory DB corrupted
• Chat sessions vanished
• Skill not loading
• Cross-machine bridge silently absent
• Logs too big (no pm2-logrotate)
• Approval gate bypass (allow-listed surface)
• Mobile PWA broken (cert expired)
• Terminal mode empty
• mc_remote_ask truncated

Plus a "diagnostic sweep" command sequence for when none of the above match, a list of key files to know, and 10 known footguns to read BEFORE editing the codebase. Run Claude Code in your fusio directory and ask "fix mc" to see it in action.

How to use it

The chat is the front door. From there you reach everything else. Full guide in the repo at docs/HELP_GUIDE.md.

Send a message

Type into the composer. Press Enter to send. Shift+Enter for newline. The agent decides what skills to load based on your prompt — you'll see them in the chip bar after sending.

Start a mission (long-running multi-agent build)

Top nav → Luke's Chat. Pick a preset (frontier-mixed for most work, full-force for ship-or-die, budget for cheap recon). Write the goal in the composer — describe the deliverable, not the steps. The orchestrator drafts a plan with typed assertions; you approve. Then it runs phase-by-phase, with parallel scrutiny lanes. Survives MC restarts. Checkpoints + rewind included.

Add a peer machine (cross-machine bridge)

Create ~/.config/mc-remote-hosts.json on each machine with symmetric bearer tokens:

{
  "myToken": "shared-token-here",
  "myLabel": "Server A",
  "myUrl": "http://10.0.0.5:3001",
  "hosts": [
    { "id": "b", "label": "Workstation B",
      "url": "http://10.0.0.6:3001",
      "token": "shared-token-here" }
  ]
}

Both machines need each other in their hosts. Once set, the Remote tab on each shows the other's MC as an embedded iframe with a reachability dot. The chat agent gets mc_remote_ask / mc_remote_read / mc_remote_recover MCP tools.

Install as a mobile PWA

Generate HTTPS certs for whatever hostname your phone will reach (Tailscale's tailscale cert, Let's Encrypt, mkcert, etc). Drop them at certs/tls.crt and certs/tls.key. Start the HTTPS proxy (node https-proxy.js or via PM2's mc-https app). Visit the HTTPS URL on your phone and Chrome will offer to install the PWA.

Add a custom skill

Drop a SKILL.md with frontmatter into one of the 5 source repos under ~/:

---
name: my-skill
description: When to use this. The agent sees this in mc_list_skills.
triggers:
  - "fix the alignment"
  - "tighten the layout"
---

# My skill body — under 5.5 KB inline cap, longer is fine but truncates.

## When to use
...

## How to apply
...

Verify with curl 'http://localhost:3001/api/debug/match-probe?text=your+trigger+phrase'. If bytesInlined > 0 for your skill, it's wired.

How it compares

Not because we want to dunk on commercial tools — because the differences are real and structural.

The kind of teams who get the most out of this

AiFusio MC was shaped by real client work across AI Fusio's target industries. Here's where it fits naturally.

Want this customized to run your business?

Cloning AiFusio MC gets you the platform. AI Fusio builds the actual integration — your CRM, your phones, your documents, your team's specific workflows. We take this same codebase and shape it around your business, train your team, and operate it as your AI department for as long as you need.

Why this is different from what's on the market

Most agentic tools fall into one of two camps. AiFusio MC is in neither — and the difference matters when you actually try to use AI for real work.

FAQ

Bite-sized answers. Full FAQ in the repo at docs/FAQ.md.