fix: add Podman support with NVIDIA CDI GPU passthrough

Makefile

@@ -1,6 +1,7 @@
 .PHONY: help build build-admin build-voice dev dev-admin dev-voice clean download-model swagger docker-build docker-buildx docker-push postgres redis ollama infra infra-stop infra-clean
 
 CONFIG ?= config.yaml
+CONTAINER_ENGINE ?= $(shell command -v podman 2>/dev/null || echo docker)
 
 help:
 	@echo "Magec - Multi-Agent AI Platform"
@@ -91,54 +92,54 @@ IMAGE_TAG ?= latest
 DOCKER_PLATFORMS ?= linux/amd64,linux/arm64
 
 docker-build:
-	@docker build -f docker/build/Dockerfile -t $(IMAGE_NAME):$(IMAGE_TAG) .
+	@$(CONTAINER_ENGINE) build -f docker/build/Dockerfile -t $(IMAGE_NAME):$(IMAGE_TAG) .
 	@echo "Image built: $(IMAGE_NAME):$(IMAGE_TAG)"
 
 docker-buildx:
-	@docker buildx build -f docker/build/Dockerfile --platform $(DOCKER_PLATFORMS) -t $(IMAGE_NAME):$(IMAGE_TAG) .
+	@$(CONTAINER_ENGINE) buildx build -f docker/build/Dockerfile --platform $(DOCKER_PLATFORMS) -t $(IMAGE_NAME):$(IMAGE_TAG) .
 	@echo "Multi-arch image built: $(IMAGE_NAME):$(IMAGE_TAG) [$(DOCKER_PLATFORMS)]"
 
 docker-push:
-	@docker buildx build -f docker/build/Dockerfile --platform $(DOCKER_PLATFORMS) -t $(IMAGE_NAME):$(IMAGE_TAG) --push .
+	@$(CONTAINER_ENGINE) buildx build -f docker/build/Dockerfile --platform $(DOCKER_PLATFORMS) -t $(IMAGE_NAME):$(IMAGE_TAG) --push .
 	@echo "Image pushed: $(IMAGE_NAME):$(IMAGE_TAG) [$(DOCKER_PLATFORMS)]"
 
 # Infrastructure (Docker)
 
 postgres:
-	@docker run -d --name magec-postgres \
+	@$(CONTAINER_ENGINE) run -d --name magec-postgres \
 		-p 5432:5432 \
 		-e POSTGRES_PASSWORD=postgres \
 		-e POSTGRES_DB=magec \
-		pgvector/pgvector:pg17
+		docker.io/pgvector/pgvector:pg17
 	@echo "PostgreSQL (pgvector) started on localhost:5432"
 
 redis:
-	@docker run -d --name magec-redis \
+	@$(CONTAINER_ENGINE) run -d --name magec-redis \
 		-p 6379:6379 \
-		redis:alpine
+		docker.io/library/redis:alpine
 	@echo "Redis started on localhost:6379"
 
 ollama:
-	@docker run -d --name magec-ollama \
+	@$(CONTAINER_ENGINE) run -d --name magec-ollama \
 		-p 11434:11434 \
 		-v ollama:/root/.ollama \
-		ollama/ollama
+		docker.io/ollama/ollama
 	@echo "Waiting for Ollama to start..."
 	@sleep 3
-	@docker exec magec-ollama ollama pull qwen3:8b
-	@docker exec magec-ollama ollama pull nomic-embed-text
+	@$(CONTAINER_ENGINE) exec magec-ollama ollama pull qwen3:8b
+	@$(CONTAINER_ENGINE) exec magec-ollama ollama pull nomic-embed-text
 	@echo "Ollama started on localhost:11434 with qwen3:8b and nomic-embed-text"
 
 infra: postgres redis
 	@echo "Infrastructure ready"
 
 infra-stop:
-	@docker stop magec-postgres magec-redis 2>/dev/null || true
-	@docker rm magec-postgres magec-redis 2>/dev/null || true
+	@$(CONTAINER_ENGINE) stop magec-postgres magec-redis 2>/dev/null || true
+	@$(CONTAINER_ENGINE) rm magec-postgres magec-redis 2>/dev/null || true
 	@echo "Infrastructure stopped"
 
 infra-clean: infra-stop
-	@docker stop magec-ollama 2>/dev/null || true
-	@docker rm magec-ollama 2>/dev/null || true
-	@docker volume rm ollama 2>/dev/null || true
+	@$(CONTAINER_ENGINE) stop magec-ollama 2>/dev/null || true
+	@$(CONTAINER_ENGINE) rm magec-ollama 2>/dev/null || true
+	@$(CONTAINER_ENGINE) volume rm ollama 2>/dev/null || true
 	@echo "All containers and volumes removed"

docker/compose/docker-compose.yaml

@@ -2,7 +2,9 @@
 # Everything runs on your machine: LLM, STT, TTS, embeddings, memory.
 # No API keys, no cloud accounts, no data leaving your network.
 #
-# GPU: Uncomment the 'deploy' section under ollama for NVIDIA acceleration.
+# GPU: Uncomment ONE of the GPU sections under ollama:
+#   - Docker:  the 'deploy' block (uses nvidia container runtime)
+#   - Podman:  the 'devices' line (uses CDI — requires nvidia-ctk cdi generate)
 #
 # Usage:
 #   docker compose up -d
@@ -31,13 +33,13 @@ services:
     restart: unless-stopped
 
   redis:
-    image: redis:alpine
+    image: docker.io/library/redis:alpine
     volumes:
       - redis_data:/data
     restart: unless-stopped
 
   postgres:
-    image: pgvector/pgvector:pg17
+    image: docker.io/pgvector/pgvector:pg17
     environment:
       POSTGRES_USER: magec
       POSTGRES_PASSWORD: magec
@@ -47,21 +49,27 @@ services:
     restart: unless-stopped
 
   ollama:
-    image: ollama/ollama:latest
+    image: docker.io/ollama/ollama:latest
     volumes:
       - ollama_data:/root/.ollama
     restart: unless-stopped
-    # Uncomment for NVIDIA GPU acceleration:
+    # ── NVIDIA GPU acceleration (uncomment ONE option) ──────────────
+    #
+    # Option A — Docker (nvidia-container-runtime):
     # deploy:
     #   resources:
     #     reservations:
     #       devices:
     #         - driver: nvidia
     #           count: all
     #           capabilities: [gpu]
+    #
+    # Option B — Podman (CDI — run: sudo nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml):
+    # devices:
+    #   - nvidia.com/gpu=all
 
   ollama-setup:
-    image: ollama/ollama:latest
+    image: docker.io/ollama/ollama:latest
     depends_on:
       - ollama
     restart: "no"
@@ -87,7 +95,7 @@ services:
     restart: unless-stopped
 
   tts:
-    image: travisvn/openai-edge-tts:latest
+    image: docker.io/travisvn/openai-edge-tts:latest
     environment:
       - REQUIRE_API_KEY=False
     restart: unless-stopped

scripts/install.sh

@@ -270,19 +270,19 @@ box_line "  on your machine."
 box_empty
 box_sep
 box_empty
-box_line "$(badge " 2 " "$BG_MAGENTA" "$FG_WHITE")  Use containers (Docker)"
+box_line "$(badge " 2 " "$BG_MAGENTA" "$FG_WHITE")  Use containers (Docker / Podman)"
 box_empty
 box_line "  Everything runs inside isolated containers."
 box_line "  Nothing gets installed system-wide. Easiest"
 box_line "  to set up — one command starts everything."
-box_line "  Requires Docker to be installed."
+box_line "  Requires Docker or Podman to be installed."
 box_empty
 box_bottom
 echo
 
 choose \
   "Download the program directly (binary)" \
-  "Use containers (Docker)"
+  "Use containers (Docker / Podman)"
 
 INSTALL_METHOD="$REPLY"
 
@@ -713,7 +713,7 @@ printf "  ${DIM}$(hline '─' "$BOX_W")${NC}\n"
 echo
 
 method_label="Direct download (binary)"
-[[ "$INSTALL_METHOD" == "2" ]] && method_label="Containers (Docker)"
+[[ "$INSTALL_METHOD" == "2" ]] && method_label="Containers (Docker/Podman)"
 
 llm_label="Local (Ollama — private)"
 [[ "$LLM_CHOICE" == "2" ]] && llm_label="Cloud (OpenAI, Anthropic, Gemini)"
@@ -1393,21 +1393,39 @@ install_containers() {
   printf "  ${DIM}$(hline '─' "$BOX_W")${NC}\n"
   echo
 
-  # ── Check Docker ────────────────────────────────────────────────────
+  # ── Check container engine (Docker or Podman) ───────────────────────
 
-  info "Checking for Docker..."
+  CONTAINER_ENGINE=""
+  USE_PODMAN=false
 
-  if ! check_cmd docker; then
+  info "Checking for a container engine..."
+
+  if check_cmd podman && podman info &>/dev/null 2>&1; then
+    CONTAINER_ENGINE="podman"
+    USE_PODMAN=true
+    ok "Podman is ready"
+  elif check_cmd docker && docker info &>/dev/null 2>&1; then
+    CONTAINER_ENGINE="docker"
+    ok "Docker is ready"
+  elif check_cmd docker; then
+    die "Docker is installed but not running. Start Docker and try again."
+  elif check_cmd podman; then
+    die "Podman is installed but not running. Start Podman and try again."
+  else
     echo
     box_top
     box_empty
-    box_line "  ${RED}Docker is not installed${NC}" "" "center"
+    box_line "  ${RED}No container engine found${NC}" "" "center"
     box_empty
-    box_line "  Docker is needed to run Magec in containers."
-    box_line "  Install it and then run this script again:"
+    box_line "  A container engine is needed to run Magec."
+    box_line "  Install Docker or Podman and try again:"
     box_empty
     case "$OS" in
-      linux)  box_line "  ${CYAN}https://docs.docker.com/engine/install/${NC}" ;;
+      linux)
+        box_line "  ${CYAN}https://docs.docker.com/engine/install/${NC}"
+        box_line "  ${DIM}or${NC}"
+        box_line "  ${CYAN}https://podman.io/docs/installation${NC}"
+        ;;
       darwin) box_line "  ${CYAN}https://docs.docker.com/desktop/install/mac-install/${NC}" ;;
       windows) box_line "  ${CYAN}https://docs.docker.com/desktop/install/windows-install/${NC}" ;;
     esac
@@ -1416,18 +1434,48 @@ install_containers() {
     exit 1
   fi
 
-  if ! docker info &>/dev/null; then
-    die "Docker is installed but not running. Start Docker and try again."
-  fi
+  # ── Check compose ───────────────────────────────────────────────────
 
-  if ! docker compose version &>/dev/null && ! docker-compose version &>/dev/null; then
-    die "Docker Compose is required. Install it from https://docs.docker.com/compose/install/"
+  if $USE_PODMAN; then
+    if podman compose version &>/dev/null 2>&1; then
+      COMPOSE="podman compose"
+    elif check_cmd podman-compose; then
+      COMPOSE="podman-compose"
+    else
+      die "podman-compose is required. Install it: pip install podman-compose  (or use podman with compose plugin)"
+    fi
+  else
+    if docker compose version &>/dev/null; then
+      COMPOSE="docker compose"
+    elif docker-compose version &>/dev/null; then
+      COMPOSE="docker-compose"
+    else
+      die "Docker Compose is required. Install it from https://docs.docker.com/compose/install/"
+    fi
   fi
 
-  ok "Docker is ready"
+  ok "Compose command: ${COMPOSE}"
+
+  # ── Check GPU / NVIDIA ─────────────────────────────────────────────
 
   if $GPU; then
-    if ! docker info 2>/dev/null | grep -qi 'nvidia'; then
+    local gpu_ok=false
+
+    if $USE_PODMAN; then
+      # Podman uses CDI — check for nvidia CDI spec
+      if [[ -f /etc/cdi/nvidia.yaml ]] || { [[ -d /etc/cdi ]] && ls /etc/cdi/nvidia*.yaml &>/dev/null 2>&1; }; then
+        gpu_ok=true
+      elif nvidia-ctk cdi list 2>/dev/null | grep -qi 'nvidia.com/gpu'; then
+        gpu_ok=true
+      fi
+    else
+      # Docker — check nvidia runtime
+      if docker info 2>/dev/null | grep -qi 'nvidia'; then
+        gpu_ok=true
+      fi
+    fi
+
+    if ! $gpu_ok; then
       echo
       box_top
       box_empty
@@ -1440,20 +1488,19 @@ install_containers() {
       box_line "  ${CYAN}https://docs.nvidia.com/datacenter/cloud-native/${NC}"
       box_line "  ${CYAN}container-toolkit/install-guide.html${NC}"
       box_empty
-      box_line "  Then restart Docker and run this script again."
+      if $USE_PODMAN; then
+        box_line "  Then generate CDI specs:"
+        box_line "  ${CYAN}sudo nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml${NC}"
+      else
+        box_line "  Then restart Docker and run this script again."
+      fi
       box_empty
       box_bottom
       exit 1
     fi
     ok "NVIDIA GPU detected"
   fi
 
-  if docker compose version &>/dev/null; then
-    COMPOSE="docker compose"
-  else
-    COMPOSE="docker-compose"
-  fi
-
   # ── Generate files ──────────────────────────────────────────────────
 
   info "Creating configuration files..."
@@ -1692,7 +1739,7 @@ generate_docker_compose() {
 
   if $WANT_REDIS; then
     services+="\n  redis:\n"
-    services+="    image: redis:alpine\n"
+    services+="    image: docker.io/library/redis:alpine\n"
     services+="    volumes:\n"
     services+="      - redis_data:/data\n"
     services+="    restart: unless-stopped\n"
@@ -1701,7 +1748,7 @@ generate_docker_compose() {
 
   if $WANT_POSTGRES; then
     services+="\n  postgres:\n"
-    services+="    image: pgvector/pgvector:pg17\n"
+    services+="    image: docker.io/pgvector/pgvector:pg17\n"
     services+="    environment:\n"
     services+="      POSTGRES_USER: magec\n"
     services+="      POSTGRES_PASSWORD: magec\n"
@@ -1714,19 +1761,26 @@ generate_docker_compose() {
 
   if $need_ollama; then
     services+="\n  ollama:\n"
-    services+="    image: ollama/ollama:latest\n"
+    services+="    image: docker.io/ollama/ollama:latest\n"
     services+="    volumes:\n"
     services+="      - ollama_data:/root/.ollama\n"
     services+="    restart: unless-stopped\n"
 
     if $GPU; then
-      services+="    deploy:\n"
-      services+="      resources:\n"
-      services+="        reservations:\n"
-      services+="          devices:\n"
-      services+="            - driver: nvidia\n"
-      services+="              count: all\n"
-      services+="              capabilities: [gpu]\n"
+      if $USE_PODMAN; then
+        # Podman uses CDI (Container Device Interface) for GPU access
+        services+="    devices:\n"
+        services+="      - nvidia.com/gpu=all\n"
+      else
+        # Docker uses nvidia container runtime via deploy block
+        services+="    deploy:\n"
+        services+="      resources:\n"
+        services+="        reservations:\n"
+        services+="          devices:\n"
+        services+="            - driver: nvidia\n"
+        services+="              count: all\n"
+        services+="              capabilities: [gpu]\n"
+      fi
     fi
 
     local models_to_pull=""
@@ -1738,7 +1792,7 @@ generate_docker_compose() {
     fi
 
     services+="\n  ollama-setup:\n"
-    services+="    image: ollama/ollama:latest\n"
+    services+="    image: docker.io/ollama/ollama:latest\n"
     services+="    depends_on:\n"
     services+="      - ollama\n"
     services+="    restart: \"no\"\n"
@@ -1765,7 +1819,7 @@ generate_docker_compose() {
     services+="    restart: unless-stopped\n"
 
     services+="\n  tts:\n"
-    services+="    image: travisvn/openai-edge-tts:latest\n"
+    services+="    image: docker.io/travisvn/openai-edge-tts:latest\n"
     services+="    environment:\n"
     services+="      - REQUIRE_API_KEY=False\n"
     services+="    restart: unless-stopped\n"