Onboarding

ghost init is an interactive wizard that configures everything your GHOST needs to run. It replaces manual config file editing with guided prompts.

What It Does

The wizard runs in 5 phases:

1. Environment detection — checks for Nix, container runtime (podman/docker), existing services, available memory
2. LLM provider — pick a provider (OpenRouter, Anthropic, Kimi, or ChatGPT OAuth), enter credentials, choose a model, and validate the connection with a real API call
3. Discord — step-by-step guide to create a bot, enter the token and your user ID, validated against the Discord API
4. Services — for each service (embeddings, web search, web fetch, document processing): choose local install, container, remote URL, or skip. Installs nix packages and generates a compose file
5. Write + launch — shows a config diff, writes config.toml + .env, bootstraps the workspace, installs systemd/launchd units, starts services, runs health checks, and sends the first Discord message

Non-Interactive Mode

Every prompt has a corresponding CLI flag. Pass all flags to skip prompts entirely — useful for automated deployments and testing:

ghost init \
  --provider openrouter \
  --api-key "$OPENROUTER_API_KEY" \
  --model "anthropic/claude-sonnet-4" \
  --context-window 200000 \
  --discord-token "$DISCORD_BOT_TOKEN" \
  --discord-user "$DISCORD_USER_ID" \
  --embeddings local \
  --search local \
  --crawl local \
  --docling local \
  --start

Partial flags work too — the wizard prompts for anything not provided.

Re-running

Run ghost init again at any time. It detects your existing configuration, offers to update or start fresh, and pre-fills prompts with current values. A diff of all changes is shown before writing.

Help During Setup

After the provider step validates successfully, type h at any text prompt to chat with an onboarding assistant. It uses your configured LLM to answer questions about services, tradeoffs, and configuration — then returns you to the wizard.