deploy: Dockerfile, Compose pro Rolle, LiteLLM, Prometheus/Alertmanager, Canary-Prober, n8n-Workflows

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

sites/knaster/registry.yaml

@@ -1,5 +1,6 @@
 # Registry für den Cluster "knaster" (SPEC-core §1).
 # Clients sprechen nur Slot-Namen an; litellm_model ist der Alias in deploy/litellm/config.yaml.
+# ops: Labels für den Healer (Portainer-Endpoint, Container, Proxmox-Node, VM-ID).
 slots:
   tier0-strong:
     tier: 0
@@ -21,6 +22,7 @@ slots:
     effective_context: 16384
     supports_lora: true
     fallbacks: [tier1-flex]
+    ops: {docker_host: gpu-main, container: vllm, pve_node: pve-ada, vmid: "201"}   # VM-IDs/Hosts: ANNAHME bis Deployment
 
   tier1-flex:
     tier: 1
@@ -32,6 +34,7 @@ slots:
     max_context: 32768
     effective_context: 16384
     supports_lora: true
+    ops: {docker_host: gpu-flex, container: vllm, pve_node: pve-ada, vmid: "202"}   # VM-IDs/Hosts: ANNAHME bis Deployment
 
   verifier-small:
     tier: 1
@@ -42,6 +45,7 @@ slots:
     base_model: qwen3-4b-q4km
     max_context: 8192
     effective_context: 6144
+    ops: {docker_host: gpu-small, container: llamacpp-chat, pve_node: pve3, vmid: "231"}   # VM-IDs/Hosts: ANNAHME bis Deployment
 
   signature-extractor:
     tier: 1
@@ -52,6 +56,7 @@ slots:
     base_model: qwen3-4b-q4km
     max_context: 8192
     effective_context: 6144
+    ops: {docker_host: gpu-small, container: llamacpp-chat, pve_node: pve3, vmid: "231"}   # VM-IDs/Hosts: ANNAHME bis Deployment
 
   embed-default:
     tier: 1
@@ -62,6 +67,7 @@ slots:
     base_model: qwen3-embedding-0.6b-q8
     max_context: 8192
     kind: embedding
+    ops: {docker_host: gpu-embed-1, container: llamacpp-embed, pve_node: pve3, vmid: "221"}   # VM-IDs/Hosts: ANNAHME bis Deployment
 
   rerank-default:
     tier: 1
@@ -72,5 +78,6 @@ slots:
     base_model: bge-reranker-v2-m3-q8
     max_context: 8192
     kind: rerank
+    ops: {docker_host: gpu-embed-1, container: llamacpp-rerank, pve_node: pve3, vmid: "221"}   # VM-IDs/Hosts: ANNAHME bis Deployment
 
 adapters: []