From 57ef5a13b1d8afe16f78547a368b5ef19c3ba3ac Mon Sep 17 00:00:00 2001
From: Preston Hunter <phunter@prestonhunter.space>
Date: Sat, 27 Dec 2025 12:43:54 -0500
Subject: [PATCH] Add AI stack

---
 stacks/ai/ai-backend.hcl    |  0
 stacks/ai/ai-backend.nomad  | 65 +++++++++++++++++++++++++++++++++++++
 stacks/ai/ai-frontend.hcl   |  0
 stacks/ai/ai-frontend.nomad | 53 ++++++++++++++++++++++++++++++
 4 files changed, 118 insertions(+)
 delete mode 100644 stacks/ai/ai-backend.hcl
 create mode 100644 stacks/ai/ai-backend.nomad
 delete mode 100644 stacks/ai/ai-frontend.hcl
 create mode 100644 stacks/ai/ai-frontend.nomad

diff --git a/stacks/ai/ai-backend.hcl b/stacks/ai/ai-backend.hcl
deleted file mode 100644
index e69de29..0000000
diff --git a/stacks/ai/ai-backend.nomad b/stacks/ai/ai-backend.nomad
new file mode 100644
index 0000000..e489b57
--- /dev/null
+++ b/stacks/ai/ai-backend.nomad
@@ -0,0 +1,65 @@
+job "ai-backend" {
+  datacenters = ["Homelab-PTECH-DC"]
+  region      = "home"
+  type        = "service"
+
+  group "ollama-group" {
+    count = 1
+
+    # Pin to P52 Laptop (eGPU Host)
+    constraint {
+      attribute = "${meta.device}"
+      value     = "p52-laptop"
+    }
+
+    network {
+      port "api" { static = 11434 }
+    }
+
+    task "ollama" {
+      driver = "podman"
+
+      env {
+        OLLAMA_HOST    = "0.0.0.0"
+        OLLAMA_ORIGINS = "*"
+        
+        # CRITICAL FOR 6900XT:
+        # This tells ROCm to treat the card like a supported Pro workstation card
+        HSA_OVERRIDE_GFX_VERSION = "10.3.0"
+      }
+
+      config {
+        # Standard image (contains ROCm libraries)
+        image = "docker.io/ollama/ollama:latest"
+        ports = ["api"]
+        
+        # Required for hardware access
+        privileged = true
+
+        # Pass the graphics hardware to the container
+        volumes = [
+          "/opt/homelab/ollama:/root/.ollama",
+          "/dev/kfd:/dev/kfd",
+          "/dev/dri:/dev/dri"
+        ]
+      }
+
+      service {
+        name = "ollama"
+        port = "api"
+        
+        check {
+          type     = "http"
+          path     = "/"
+          interval = "20s"
+          timeout  = "2s"
+        }
+      }
+      
+      resources {
+        cpu    = 2000
+        memory = 8192 # 8GB System RAM (The GPU has its own VRAM)
+      }
+    }
+  }
+}
\ No newline at end of file
diff --git a/stacks/ai/ai-frontend.hcl b/stacks/ai/ai-frontend.hcl
deleted file mode 100644
index e69de29..0000000
diff --git a/stacks/ai/ai-frontend.nomad b/stacks/ai/ai-frontend.nomad
new file mode 100644
index 0000000..5ec6d71
--- /dev/null
+++ b/stacks/ai/ai-frontend.nomad
@@ -0,0 +1,53 @@
+job "ai-frontend" {
+  datacenters = ["Homelab-PTECH-DC"]
+  region      = "home"
+  type        = "service"
+
+  group "openwebui" {
+    count = 1
+    network {
+      port "http" { to = 8080 }
+    }
+    service {
+      name = "openwebui"
+      port = "http"
+      tags = ["traefik.enable=true"]
+    }
+    task "server" {
+      driver = "podman"
+      env {
+        OLLAMA_BASE_URL = "http://ollama.service.consul:11434"
+      }
+      config {
+        image = "ghcr.io/open-webui/open-webui:main"
+        ports = ["http"]
+        volumes = ["/opt/homelab/openwebui:/app/backend/data"]
+      }
+      resources { cpu = 1000; memory = 1024 }
+    }
+  }
+
+  group "lobechat" {
+    count = 1
+    network {
+      port "http" { to = 3210 }
+    }
+    service {
+      name = "lobechat"
+      port = "http"
+      tags = ["traefik.enable=true"]
+    }
+    task "server" {
+      driver = "podman"
+      env {
+        OLLAMA_PROXY_URL = "http://ollama.service.consul:11434"
+        ACCESS_CODE      = "securepassword123"
+      }
+      config {
+        image = "docker.io/lobehub/lobe-chat"
+        ports = ["http"]
+      }
+      resources { cpu = 500; memory = 512 }
+    }
+  }
+}
\ No newline at end of file