Installation

Erebros runs on your machine. There is no hosted tier and no account — you unpack or clone it, provide the credentials for the models and agents you want, and start it.

The golden rule: check before you install

Every dependency below comes with a check command. Run it first; install only when the check fails. A machine may already have Node, Ollama, Hermes or Chromium, and reinstalling risks version conflicts.

command -v <tool>      # prints a path and exits 0 if installed → skip it

Erebros is built around this: every app degrades gracefully. If an app’s external tool is missing, that app shows a setup card with instructions — the rest of the dashboard works. Install Erebros first; add tools only for the apps you actually use.

Requirements

• Node.js ≥ 20 (tested on v24) and npm — the only hard requirement.
• git if you install from the repository (self-update pulls from it).
• python3 for the skill-security gate and several engines.
• A shell. Linux (systemd), macOS (launchd) and Windows (Scheduled Task) each have their own autostart installer.

No system Node and no admin rights? Install a user-local copy:

curl -fSL https://nodejs.org/dist/v24.18.0/node-v24.18.0-$(uname -s | tr A-Z a-z)-$(uname -m | sed 's/x86_64/x64/;s/aarch64/arm64/').tar.gz -o /tmp/node.tar.gz
mkdir -p ~/.local && tar -xzf /tmp/node.tar.gz -C ~/.local && mv ~/.local/node-v24* ~/.local/node
export PATH="$HOME/.local/node/bin:$PATH"     # add this line to your shell profile too

Get the code

From a release zip (the normal path for a new machine):

unzip erebros-1.0.42.zip -d ~/erebros && cd ~/erebros

Or from the repository, which keeps the in-app updater available:

git clone <repository-url> ~/erebros && cd ~/erebros

The repository is private; cloning needs a GitHub token with read access. An install without a git repository cannot pull updates; it is updated by unpacking a newer zip.

Linux

cd ~/erebros
npm install
cp .env.example .env.local                 # then edit — every key is optional
bash scripts/setup-skill-validator.sh      # the skill-security gate (needs python3)
npm run dev                                # → http://localhost:3300

For a long-lived machine, install it as a service. The installer builds once, registers a systemd user unit (erebros.service, restart on crash, a two-minute health watchdog) and provisions the free offline engines:

bash scripts/install-autostart.sh --now
loginctl enable-linger "$USER"             # or the service stops when you log out

After that, do not run npm run dev by hand — the service owns port 3300. Restart it with systemctl --user restart erebros; read logs with journalctl --user -u erebros -f.

macOS

Same first four commands as Linux. Then either run npm run dev in a terminal, or register a launchd agent:

bash scripts/install-autostart-macos.sh --now

It writes a LaunchAgent that starts Erebros at login and restarts it on crash. Restart with launchctl kickstart -k gui/$(id -u)/com.erebros.mission-control; logs land in ~/Library/Logs/.

Windows

Use PowerShell (or Git Bash for the Unix-style commands in the other guides):

cd ~\erebros
npm install
Copy-Item .env.example .env.local
npm run dev                                # → http://localhost:3300

For autostart, run scripts/install-autostart-windows.ps1 once — it provisions the same engines and registers a Scheduled Task named Erebros that starts at logon. Stop and start it with Stop-ScheduledTask -TaskName Erebros / Start-ScheduledTask -TaskName Erebros.

Credentials

Copy the example environment file and fill in only the keys you need. Nothing is required to boot; each provider turns on when its key is present.

cp .env.example .env.local

Environment files are read only at server startup — restart after editing. Keys can also be added from Settings → Secrets inside the app, which writes the same file. See Configuration for what goes where, and Add an API key for the step-by-step.

Optional tools

Each app says what it needs on its own page. The short list, with the check:

App(s)	Tool	Check
Chat (Claude engine) · Loop · Pipeline · SEO-Forge · M-Forge Compose / OpenMontage · Mastermind	Claude Code CLI	command -v claude
Chat (Codex)	OpenAI Codex CLI	command -v codex
Chat (Grok)	Grok Build CLI	command -v grok
Chat (Antigravity)	agy CLI or the Antigravity proxy	command -v agy
Chat (Qwen Code)	qwen CLI	command -v qwen
Chat (Local) · Loop local judge	Ollama + a pulled model	command -v ollama
Hermes · Kanban · Hermes Swarm · Game Studio	Hermes Agent	command -v hermes
OpenClaw	OpenClaw	command -v openclaw
OpenCode	OpenCode CLI	command -v opencode
P-Forge courses · M-Forge Studio	python3, ffmpeg, rsvg-convert	command -v python3 ffmpeg rsvg-convert
G-Forge · P-Forge renders	Blender + Godot (portable)	blender --version
Open Design	the open-design daemon	curl -s 127.0.0.1:7456/api/health
Paperclip	the Paperclip app on :3100	curl -s 127.0.0.1:3100

scripts/setup-free-engines.sh provisions the free offline engines (Kokoro, Supertonic, Whisper, yt-dlp); scripts/setup-3d-engines.sh the 3D ones; the autostart installers run both. The full matrix is DEPENDENCIES.md in the repo.

PATH for spawned CLIs. Erebros spawns agents with PATH extended by ~/.local/bin, ~/.local/node/bin, /opt/homebrew/bin and /usr/local/bin. A tool that lives elsewhere: symlink it into ~/.local/bin.

Deploying on a server

It works, with caveats: Erebros spawns CLIs on the host and its only authentication is the remote-access PIN. Install the agents on that server, log Claude in with claude setup-token, keep the bind on loopback and reach it over Tailscale or an SSH tunnel — never expose port 3300 to the internet. Read Security & remote access before you open it to any other host.

Next

• Quickstart — first chat, the global model, first Mastermind.
• Configuration — where state and keys live.
• Updating — moving an install to a newer version in place.