Expose profile context metadata for llama.cpp model discovery

endpoint.py

@@ -10,6 +10,7 @@ import re
 import socket
 import threading
 import time
+from urllib.parse import urlsplit,parse_qs
 from http.server import BaseHTTPRequestHandler,ThreadingHTTPServer
 from api_compat import normalize_chat,CompatibilityError
 from profiles import generation_parameters
@@ -151,7 +152,29 @@ class Endpoint:
         if not row['runnable']:raise APIError('Profil derzeit nicht ausführbar: '+' '.join(row['blockers']),503,'model_unavailable')
         return row
     def model_list(self,kind="chat"):
-        return dict(object='list',data=[dict(id='athena-image' if kind=='image' else p['name'],object='model',created=int(p['updated_at']),owned_by='athena-deck') for p in self.rows() if p['enabled'] and p['runnable'] and p['kind']==kind])
+        data=[]
+        for p in self.rows():
+            if not (p['enabled'] and p['runnable'] and p['kind']==kind):continue
+            row=dict(id='athena-image' if kind=='image' else p['name'],object='model',created=int(p['updated_at']),owned_by='athena-deck')
+            if kind=='chat':
+                params=p.get('parameters',{})
+                row.update(context_window=int(params.get('context',8192)),parallel_slots=int(params.get('slots',1)),context_scope='shared')
+            data.append(row)
+        return dict(object='list',data=data)
+    def model_properties(self,name):
+        # Discovery is configuration-only: never start or probe a model worker.
+        p=self.find_profile(name,'chat')
+        params=p.get('parameters',{})
+        model=p.get('model') or {}
+        identity=' '.join(str(model.get(k,'')) for k in ('repo','file','filename','name')).lower()
+        known_template=bool(re.search(r'qwen|gemma',identity))
+        return dict(model=p['name'],n_ctx=int(params.get('context',8192)),
+                    default_generation_settings=dict(n_ctx=int(params.get('context',8192)),params={}),
+                    total_slots=int(params.get('slots',1)),context_scope='shared',
+                    modalities=dict(vision=bool(params.get('vision_projector'))),
+                    chat_template_caps=dict(supports_tool_calls=known_template,
+                                            supports_typed_content=True,
+                                            supports_reasoning_effort='qwen' in identity))
 
 class APIHTTPServer(ThreadingHTTPServer):
     daemon_threads=True
@@ -206,6 +229,16 @@ class APIHandler(BaseHTTPRequestHandler):
             with ep.lock:
                 if not ep.allowed():raise APIError('Endpunkt wird gestoppt.',503,'endpoint_stopping')
                 ep.inflight+=1;admitted=True
+            parsed=urlsplit(self.path)
+            if self.command=='GET' and (parsed.path=='/models' or (parsed.path=='/v1/models' and ep.scheduler.gpu_mode!='video')):
+                return self.send(ep.model_list())
+            if self.command=='GET' and parsed.path=='/props':
+                query=parse_qs(parsed.query)
+                name=query.get('model',[None])[0]
+                if not name:raise APIError('model ist für Router-Eigenschaften erforderlich.')
+                return self.send(ep.model_properties(name))
+            if self.command=='GET' and parsed.path=='/health':
+                return self.send({'status':'ok','service':'athena-deck-api'})
             if ep.scheduler.gpu_mode=='separator' and self.path not in ('/v1/embeddings','/v1/embeddings/models'):raise APIError('Audio-Trennung aktiv. Für Chat, Bilder oder Musik die Betriebsart wechseln.',503,'separator_mode_active')
             if ep.video and ep.scheduler.gpu_mode not in ('llm','music') and self.path not in ('/v1/embeddings','/v1/embeddings/models'):
                 if ep.scheduler.gpu_mode=='switching':raise APIError('Moduswechsel läuft.',503,'mode_switching')