Setup

Setup (step by step)

End-to-end first-run path. Prefer the guided wizard if you want checkboxes; use the manual steps when automating.

Watch it: Demo: Quickstart

---

1. Install the CLI

Pick one:

# Homebrew (macOS / Linux)
brew install --cask openhat-security/tap/runhug

# npm
npm install -g runhug

# curl installer
curl -fsSL https://raw.githubusercontent.com/openhat-security/runhug/main/scripts/install.sh | bash

Check:

runhug version    # expect 0.4.1+
runhug            # grouped help

---

2. Guided setup (recommended)

runhug wizard

Non-interactive checklist:

runhug wizard --yes

The wizard walks credentials, index packs, and an optional first search.

---

3. Credentials (config)

Tokens are not on the root help screen — they live under config:

runhug config --help

# RunPod (deploy / heretic)
runhug config connect
# or alias: runhug connect

# Hugging Face (gated models / uploads)
runhug config connect hf
# or alias: runhug connect hf

Env vars also work: RUNPOD_API_KEY, HF_TOKEN.

Inspect status (paths only — secrets never printed):

runhug config

Clear later:

runhug config disconnect
runhug config disconnect hf

---

4. Search index packs

Packs are SQLite category indexes from hfpacks Releases (not from runhug releases).

# Interactive install (default: missing packs)
runhug packs install

# Or by id / alias
runhug packs install llm gguf

# See local vs remote
runhug packs list

# Refresh installed packs after a new hfpacks release
runhug packs update

Shortcut used by wizard/init:

runhug init --yes

---

5. GPU preference (optional)

runhug gpu                 # help (not a catalog dump)
runhug gpu list --filter runpod --sort value --limit 10
runhug gpu set             # interactive preference for deploy / local

---

6. First search & inspect

runhug search -q "small instruct" --limit 10
runhug recommend "small instruct for a laptop" --no-llm --candidates 5
runhug inspect Qwen/Qwen2.5-0.5B-Instruct

Add --online / --hub only when you want live Hub results (token helps rate limits).

---

7. Dry-run deploy (no spend)

runhug deploy Qwen/Qwen2.5-0.5B-Instruct --dry-run --estimate

When ready for real deploy (needs RunPod key):

runhug deploy Qwen/Qwen2.5-0.5B-Instruct
runhug proxy               # OpenAI-compatible @ 127.0.0.1:8080/v1
runhug run                 # chat REPL (fullscreen; /mini for scrollback)
runhug start claude        # or: start opencode

---

8. Keep current

runhug upgrade             # CLI binary (brew/npm/…)
runhug packs update        # index packs from hfpacks
# or: runhug update --packs

---

Next demos

Flow	Page
Slow first-run story	Demo: Quickstart
Search / packs / recommend	Demo: Search
Deploy dry-run + GCP	Demo: Deploy
Heretic abliteration	Demo: Heretic
Built-in chat REPL	Demo: Chat
Claude Code / OpenCode	Demo: Agents
Everything in order	Demo: Full