neuronetz-gateway/demo.sh

#!/usr/bin/env bash
#
# demo.sh — the neuronetz-gateway one-command presentation.
#
# Brings up the demo stack (postgres + redis + mock-ollama + gateway) with NO
# GPU and NO model downloads, creates a demo tenant + API key via the bootstrap
# CLI *inside the gateway container*, and prints a clean summary with the key,
# the playground URL, and ready-to-paste curl commands.
#
# Usage:
#   ./demo.sh          # build + start, bootstrap a tenant/key, print summary
#   ./demo.sh --down   # tear the whole stack down (and remove volumes)
#   ./demo.sh --help   # this help
#
# Re-runnable: existing tenant/key are handled gracefully. The full API key is
# only ever printed once at creation (SPEC §11), so on a re-run where the key
# already exists this script creates a fresh, uniquely-named key and prints it.
set -euo pipefail

# ──────────────────────────────────────────────────────────────────────────
# Configuration
# ──────────────────────────────────────────────────────────────────────────
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" >/dev/null 2>&1 && pwd)"
COMPOSE_FILE="${SCRIPT_DIR}/docker-compose.demo.yml"
COMPOSE=(docker compose -f "${COMPOSE_FILE}")

GATEWAY_URL="http://localhost:8080"
PLAYGROUND_URL="${GATEWAY_URL}/playground"
TENANT_NAME="demo"
KEY_NAME="demo-key"

# Colours (disabled when stdout is not a TTY).
if [ -t 1 ]; then
  BOLD="$(printf '\033[1m')"; DIM="$(printf '\033[2m')"; RESET="$(printf '\033[0m')"
  CYAN="$(printf '\033[36m')"; GREEN="$(printf '\033[32m')"; YELLOW="$(printf '\033[33m')"
else
  BOLD=""; DIM=""; RESET=""; CYAN=""; GREEN=""; YELLOW=""
fi

log()  { printf '%s\n' "${CYAN}==>${RESET} ${BOLD}$*${RESET}"; }
warn() { printf '%s\n' "${YELLOW}!!${RESET} $*" >&2; }
die()  { printf '%s\n' "${YELLOW}xx${RESET} $*" >&2; exit 1; }

# ──────────────────────────────────────────────────────────────────────────
# Subcommands
# ──────────────────────────────────────────────────────────────────────────
usage() {
  sed -n '3,18p' "${BASH_SOURCE[0]}" | sed 's/^# \{0,1\}//'
}

down() {
  log "Tearing down the demo stack (containers + volumes)…"
  "${COMPOSE[@]}" down --volumes --remove-orphans
  log "Done. The demo stack is gone."
}

# Run the bootstrap CLI inside the running gateway container.
gw_cli() {
  "${COMPOSE[@]}" exec -T gateway neuronetz-gateway "$@"
}

wait_for_health() {
  log "Waiting for the gateway to become healthy at ${GATEWAY_URL}/healthz …"
  local deadline=$(( $(date +%s) + 180 ))
  until curl -fsS "${GATEWAY_URL}/healthz" >/dev/null 2>&1; do
    if [ "$(date +%s)" -ge "${deadline}" ]; then
      warn "Gateway did not become healthy in time. Recent gateway logs:"
      "${COMPOSE[@]}" logs --tail=50 gateway >&2 || true
      die "Aborting."
    fi
    sleep 2
  done
  log "Gateway is up."
}

# Create the demo tenant if it does not already exist (idempotent).
ensure_tenant() {
  log "Creating demo tenant '${TENANT_NAME}' (allow-all-models) …"
  local out
  if out="$(gw_cli create-tenant --name "${TENANT_NAME}" --allow-all-models 2>&1)"; then
    printf '%s\n' "${DIM}${out}${RESET}"
  else
    # Already-exists (or similar) is fine — surface it but keep going.
    if printf '%s' "${out}" | grep -qiE 'exist|duplicate|unique'; then
      log "Tenant '${TENANT_NAME}' already exists — reusing it."
    else
      warn "create-tenant reported:"
      printf '%s\n' "${out}" >&2
      warn "Continuing; the tenant may already be present."
    fi
  fi
}

# Create a fresh API key and capture the printed key. The key is printed once.
# We give each created key a unique name so re-runs always succeed and always
# yield a usable key to print.
create_key() {
  local unique_name="${KEY_NAME}-$(date +%Y%m%d-%H%M%S)"
  log "Creating API key '${unique_name}' for tenant '${TENANT_NAME}' …" >&2
  local out
  if ! out="$(gw_cli create-key --tenant "${TENANT_NAME}" --name "${unique_name}" 2>&1)"; then
    warn "create-key failed:" >&2
    printf '%s\n' "${out}" >&2
    return 1
  fi
  # The CLI prints both the 12-char prefix (e.g. "prefix nz_abc12345Yz") AND the
  # full key on a later line. Both match /nz_[A-Za-z0-9]+/, so pick the longest
  # match — that's the full key (44 chars), never the prefix (12).
  local key
  key="$(printf '%s' "${out}" | grep -oE 'nz_[A-Za-z0-9]+' \
       | awk '{ if (length($0) > maxlen) { maxlen = length($0); k = $0 } } END { print k }' \
       || true)"
  if [ -z "${key}" ]; then
    warn "Could not parse an API key from create-key output:" >&2
    printf '%s\n' "${out}" >&2
    return 1
  fi
  printf '%s' "${key}"
}

print_summary() {
  local key="$1"
  local cl='application/json'

  cat <<EOF

${GREEN}${BOLD}════════════════════════════════════════════════════════════════════════${RESET}
${GREEN}${BOLD}  neuronetz-gateway demo is live${RESET}
${GREEN}${BOLD}════════════════════════════════════════════════════════════════════════${RESET}

  ${BOLD}API base URL${RESET}     ${CYAN}${GATEWAY_URL}${RESET}
  ${BOLD}Playground${RESET}       ${CYAN}${PLAYGROUND_URL}${RESET}
  ${BOLD}API key${RESET}          ${YELLOW}${key}${RESET}
                   ${DIM}(printed once — copy it now; re-run ./demo.sh to mint another)${RESET}

  ${BOLD}Model backend${RESET}    mock-ollama (internal network only, never published)
  ${BOLD}Models${RESET}           llama3.1:8b · mistral:7b · qwen2.5:3b · nomic-embed-text

${BOLD}── Ready-to-paste curl commands ───────────────────────────────────────${RESET}

${DIM}# 1) Streaming chat — OpenAI-compatible SSE (data: {...}  …  data: [DONE])${RESET}
curl -N ${GATEWAY_URL}/v1/chat/completions \\
  -H "Authorization: Bearer ${key}" \\
  -H "Content-Type: ${cl}" \\
  -d '{"model":"llama3.1:8b","stream":true,"messages":[{"role":"user","content":"Say hello in one sentence."}]}'

${DIM}# 2) Streaming chat — native Ollama NDJSON (one JSON object per line)${RESET}
curl -N ${GATEWAY_URL}/api/chat \\
  -H "Authorization: Bearer ${key}" \\
  -H "Content-Type: ${cl}" \\
  -d '{"model":"llama3.1:8b","stream":true,"messages":[{"role":"user","content":"Say hello in one sentence."}]}'

${DIM}# 3) List models — the tenant's effective (live-discovered) set, OpenAI format${RESET}
curl ${GATEWAY_URL}/v1/models \\
  -H "Authorization: Bearer ${key}"

${DIM}# 4) 401 Unauthorized — no/invalid key, fail-closed, no upstream details leaked${RESET}
curl -i ${GATEWAY_URL}/v1/models \\
  -H "Authorization: Bearer nz_invalidKEYdoesNotExist000000000000000000"

${DIM}# 5) 403 Forbidden — model-mutating endpoint is hard-blocked (not configurable)${RESET}
curl -i ${GATEWAY_URL}/api/pull \\
  -H "Authorization: Bearer ${key}" \\
  -H "Content-Type: ${cl}" \\
  -d '{"model":"llama3.1:8b"}'

${BOLD}───────────────────────────────────────────────────────────────────────${RESET}
  Tear it all down with:  ${CYAN}./demo.sh --down${RESET}

EOF
}

up() {
  command -v docker >/dev/null 2>&1 || die "docker is required but not found on PATH."
  command -v curl   >/dev/null 2>&1 || die "curl is required but not found on PATH."
  [ -f "${COMPOSE_FILE}" ] || die "Missing ${COMPOSE_FILE}"

  log "Building and starting the demo stack (postgres + redis + mock-ollama + gateway) …"
  "${COMPOSE[@]}" up --build -d

  wait_for_health
  ensure_tenant

  local key
  if ! key="$(create_key)"; then
    die "Could not create/parse an API key. See logs above."
  fi

  print_summary "${key}"
}

# ──────────────────────────────────────────────────────────────────────────
# Entry point
# ──────────────────────────────────────────────────────────────────────────
main() {
  case "${1:-}" in
    --down|-d|down)   down ;;
    --help|-h|help)   usage ;;
    "")               up ;;
    *)                die "Unknown argument: $1 (try --help)" ;;
  esac
}

main "$@"