Nova Cortex — multi-agent AI workforce for your business. Built on cortextOS.
Multi-agent AI workforce for your business. A curated team of always-on agents — Orchestrator + Analyst — controlled from Telegram or Slack, running on your own machine or server. Mac/Linux installs can run on Codex/OpenAI (codex-app-server) or Claude Code; specialist agents (CFO, marketer, ops, research, storytelling, copywriting, anything you need) are spawned once the core is online.
Built on top of the cortextOS engine (open-source multi-agent framework, MIT).
Specialist agents (anything domain-specific you need) are added later — the Orchestrator helps you spawn them on demand once the core pair is online. Nova Academy’s course walks through this.
Pentru curs, instalarea are trei decizii:
Ghidul complet este in docs/installation-options.md. Pentru Slack, foloseste si docs/slack-onboarding.md. Pentru validarea pe Windows, foloseste docs/windows-test-plan.md.
Mac or Linux (Ubuntu/Debian) terminal:
git clone https://github.com/<your-fork>/nova-agents.git
cd nova-agents
bash nova-init.sh
Windows PowerShell (native — not WSL):
# Open PowerShell as Administrator (required — see below)
git clone https://github.com/<your-fork>/nova-agents.git
cd nova-agents
.\nova-prereq.ps1 # installs VS Build Tools + jq + Python + Node + Claude + cortextOS + PM2
.\nova-init.ps1 # wizard (can run from regular PowerShell after prereq passes)
PowerShell must be Administrator when running
nova-prereq.ps1— winget needs admin rights to install Visual Studio Build Tools, jq, and Python globally. After prereq,nova-init.ps1runs fine from a regular PowerShell. The script halts with clear instructions if started without admin.
WSL2 is not supported. Nova Cortex runs natively on Windows via PowerShell — the cortextOS engine and Claude Code’s PTY have edge cases under WSL that don’t have clean fixes.
The Mac/Linux wizard will:
codex or claude), cortextOS engine, PM2 (and Homebrew on Mac, jq on Linux).The Windows PowerShell wizard currently follows the original Telegram flow.
For Telegram, you’ll need two BotFather tokens total: one for the Orchestrator (asked here), one for the Analyst (asked later by the Orchestrator during /onboarding). Create both ahead of time from @BotFather on Telegram if you want a smooth flow.
For Slack, create a Slack app with Socket Mode enabled before running the wizard. You need:
xoxb-...xapp-... with connections:writeC123...app_mentions:read, channels:history, chat:write, files:read, im:history, im:readapp_mention, message.channels, message.imAfter changing scopes or events in Slack, click Reinstall to Workspace. Without reinstalling, Slack will not deliver the new event types. The bridge listens to normal messages only in SLACK_LISTEN_CHANNELS; elsewhere, mention the app explicitly.
For Codex/OpenAI, run codex interactively once to sign in with ChatGPT/OpenAI, or set OPENAI_API_KEY for the user running the agents. The wizard reminds you if this is missing.
For Claude Code, run claude interactively once to log in to Claude Code. The wizard reminds you if this is missing.
Open your chosen channel:
hello.Then send:
/onboarding
The Orchestrator will walk you through identity, working hours, autonomy rules, daily goal cascade — and then ask you for the second BotFather token to bring the Analyst online. After that, you have a working 2-agent team and the Orchestrator can help you add specialist agents whenever you’re ready.
If you ever need to manually restart the agent:
# Mac/Linux:
cd ~/cortextos && cortextos start boss
# Windows PowerShell:
cd $env:USERPROFILE\cortextos; cortextos start boss
| Path | Purpose |
|---|---|
nova-prereq.sh |
Mac/Linux prereq checker — auto-installs Homebrew (Mac), jq, Node 20+, Claude Code, cortextOS, PM2. Idempotent. |
nova-prereq.ps1 |
Windows-native equivalent (PowerShell 5.1+). Installs Node via winget, Claude Code + cortextOS + PM2 via npm. Idempotent. |
nova-init.sh |
Mac/Linux student wizard. Picks runtime + workspace name + Telegram or Slack, then provisions the Orchestrator. |
templates/nova-cortex-orchestrator-codex/ |
Codex/OpenAI Orchestrator template (runtime: codex-app-server, model: gpt-5-codex). |
templates/nova-cortex-analyst-codex/ |
Codex/OpenAI Analyst template (runtime: codex-app-server, model: gpt-5-codex). |
slack-bridge/ |
Optional Slack Socket Mode bridge. Forwards Slack messages into the cortextOS bus and posts agent replies back to Slack. |
nova-init.ps1 |
Windows-native wizard (PowerShell). Original Telegram setup flow. |
templates/nova-cortex-orchestrator/ |
Branded Orchestrator template — installed into $HOME/cortextos/templates/ by either init script. |
templates/nova-cortex-analyst/ |
Branded Analyst template — installed alongside, spawned by the Orchestrator during /onboarding. |
LICENSE |
MIT, with cortextOS attribution. |
Want to fork this for your own brand?
templates/nova-cortex-* directory to templates/<your-brand>-*.IDENTITY.md, the CLAUDE.md header, the boot message in ONBOARDING.md Step 1, and the identity template in Step 18 (orchestrator only).nova-init.sh to install your renamed templates + adjust the wizard text.ONBOARDING.md Step 26 to spawn your renamed analyst template (--template <your-brand>-analyst).LICENSE attribution to your own name (keep the cortextOS notice — that’s the MIT obligation).Nova Cortex is a branding + curated-templates layer on top of cortextOS. The actual multi-agent runtime, daemon, bus, knowledge base, and Telegram integration all come from cortextOS — an open-source framework by Cortext LLC (MIT licensed). Nova’s Slack option is implemented as a lightweight bridge on top of the cortextOS bus, so it can stay compatible with upstream cortextOS.
We don’t fork cortextOS. We pin to its releases and ship our templates on top. That means cortextOS updates flow downstream automatically.
This is the bootstrap repo for the Nova Academy course on agent orchestration. The Orchestrator + Analyst templates are production-grade and identical in capability to the cortextOS originals; the Nova Cortex layer adds brand and wraps the install in a friendly wizard. Specialist agents are intentionally not pre-built — the course teaches you how to spawn them yourself.
Issues, ideas, PRs welcome.