A local command center that runs Claude Code, Hermes, Codex and a dozen other agent CLIs from one screen, with voice, memory, SEO, Shorts, Brand Studio and 3D Worlds built in. It runs on your machine at 127.0.0.1:3737 and your keys stay in your own files.
What you build
| Part | What it is | What it does |
|---|---|---|
| The dashboard | A Next.js app on 127.0.0.1:3737, built and run with Node 22 on Mac, Linux or Windows through WSL2 | One sidebar for every tool. Nothing is exposed to your network; settings live in ~/.aiab-os/config.json. |
| The agent tabs | Chat tabs for Claude Code, Hermes, Codex, OpenCode, Antigravity, Kimi, Grok, Muse and more, each on the CLI you installed | Same chat window for every agent. Long threads are trimmed newest-first so the agent keeps your latest messages. |
| Memory | Your Obsidian vault, wired in as a 3D Memory Galaxy and as Jarvis memory | Agents read and write notes in your vault, so what you decided last week is still there. |
| Voice | Hermes Apollo, a voice console on the Codex CLI with your ChatGPT login, or your OpenAI key | Talk to your agents out loud, and share a screen so it can answer about what you are looking at. |
| The studios | SEO pipeline and Sites, Shorts Studio, Thumbnail, Video, Music and 3D Worlds tabs | Write, edit and publish from the same place. Nothing posts until you press Post. |
| The add-on prompts | The eleven prompts Brand Studio, SEO Office, Software Factory and the WSL install were built with, plus a final audit | Brand Studio ships built in. SEO Office and Software Factory need their own open-source service, which the prompts install. Rebuild or adapt any of them. |
Where this comes from
AI Automation Base OS started as a community agent-dashboard pack, used here with its author's permission. We rebranded it, moved its settings to ~/.aiab-os, and gave it the AI Auto Base theme. Then we added what we kept wishing it had: Brand Studio, an SEO Office tab, a Software Factory tab, a Herdr terminal tab, real tool logos in place of emoji, and a tested install path for Windows through WSL2. It is kept in sync with upstream fixes. The latest sync, 15 October 2026, brought Shorts Studio, SEO Sites, the Antigravity model picker and the chat-history trimming.
This pack was made with inspiration from the work of:
- Andy Haffel, AI Mate
- Julian Goldie, AI Profit Lab
- Eric Tech, Eric Tech
- Jake Van Clief, Clief Notes
- AndyNoCode, AndyNoCode
- Chase Hannegan, Chase AI Community
- and others in the AI builder community whose public work shaped this one.
Two of the add-ons wrap open-source projects that stay in their own folders and keep their own licences: SEO Office is seo-os by AgriciDaniel (AGPL-3.0), running as a separate local service. Software Factory drives super-board by Eric Tech (MIT). CREDITS.md in the zip lists them.
What you build
One dashboard on 127.0.0.1:3737 with a sidebar for every agent and studio you have installed. A tab whose tool is missing tells you which install guide adds it, and Check My Setup shows the whole list. You add pieces one at a time, and every piece runs on your machine with your own keys.
The one thing it deliberately does not do: it never listens on your network. It binds to localhost, and nothing posts, publishes or deploys until you press the button.
Do you need this? Be honest first
This is for someone who already runs two or three agent CLIs and is tired of a terminal per tool. If you use one agent in one editor, keep doing that. This adds a Node app, a config file and a service to look after.
It is also big. More than fifty install guides, most of them optional. If you want a single focused system, one of the smaller packs is a better first build. Come back to this one when you have outgrown a single tab.
What you need
- A computer. A Mac, Linux, or Windows 10/11 with WSL2. The app needs about 300 MB, plus
npm install, plus whatever the studios download. - Node 22 or newer. The app uses
node:sqlite, and Node 20 fails the build. - At least one agent: Claude Code on your own plan, Codex on your ChatGPT login, Hermes, or one of the free routes in
install/5-FREE-CLAUDE-CODE.mdandinstall/28-OMNIROUTE.md. Everything else is optional. - Time. The core dashboard takes 30 to 60 minutes. Each studio after that takes 10 to 40.
What is in the zip
aiab-os/
README.md start here: the three install paths and the list of guides
IMPLEMENTATION-CHECKLIST tick-box list, only the first four steps are required
SETUP-WITH-AI.md the playbook an agent follows to install it for you
SETUP-GUIDE.md connect models, tools and keys
DISCLAIMER.md as is, no warranty, use at your own risk
CREDITS.md the open-source projects it connects to and their licences
CHANGELOG.md what changed, by day
install/ one guide per tab or studio (0 to 55)
prompts/ WSL install, Brand Studio B0-B5, SEO Office, Software Factory, Worlds check, audit
source/ the Next.js app
tools/wsl-sync-build.ps1 copy, build and restart on WSL in one command
The second zip, aiab-os-world-sources.zip, holds the editable Blender files for the 3D Worlds. You only need it to change those scenes. The worlds run without it.
What was verified and what was not
Verified on Windows 11 with WSL2 Debian 13 and Node 22.22.3, on 15 October 2026. The production build is clean. The app runs as a systemd user service. /, /claude, /seo, /opencode, /antigravity and the new Shorts, Sites and Antigravity model endpoints answer 200. The SEO regression passes 98 assertions. The chat-history and Muse self-checks pass. On 10 October, Brand Studio passed 43 tests, the icon set 38, SEO Office 5 and Software Factory 7.
Not verified: a fresh install on a Mac or on bare Linux. Agent chats against live Claude and Codex logins. Shorts posting through a real Blotato account. Sites against a real website folder. Apollo voice. Brand Studio with paid providers (only its Demo providers were run). Treat those as untested until your own machine says otherwise.
Read the prompt before you run it
Hand each prompt to your agent first and ask what it plans to change. Compare that to what the prompt says it builds. The same goes for SETUP-WITH-AI.md: let the agent read it and tell you the steps before it runs any of them. A prompt costs a minute to fix. A wrong install costs an evening.
