feat: add llama-swap as a backend

backends/normalize.py

@@ -50,27 +50,46 @@ def dedupe_on_keys(dicts, key_fields):
     return out
 
 
+def is_llama_swap(endpoint: str) -> bool:
+    """True if the endpoint is a configured llama-swap front."""
+    return endpoint in get_config().llama_swap_endpoints
+
+
+def is_llama_server(endpoint: str) -> bool:
+    """True for a llama.cpp llama-server OR a llama-swap front.
+
+    Both speak the same OpenAI-compatible surface, so the router treats them
+    identically everywhere except loaded-model detection and model unload.
+    """
+    cfg = get_config()
+    return endpoint in cfg.llama_server_endpoints or endpoint in cfg.llama_swap_endpoints
+
+
+def llama_endpoints(cfg) -> list:
+    """Combined, de-duplicated llama-server + llama-swap endpoints (order preserved)."""
+    return list(dict.fromkeys([*cfg.llama_server_endpoints, *cfg.llama_swap_endpoints]))
+
+
 def is_ext_openai_endpoint(endpoint: str) -> bool:
     """
-    Determine if an endpoint is an external OpenAI-compatible endpoint (not Ollama or llama-server).
+    Determine if an endpoint is an external OpenAI-compatible endpoint (not Ollama, llama-server or llama-swap).
 
     Returns True for:
     - External services like OpenAI.com, Groq, etc.
 
     Returns False for:
     - Ollama endpoints (without /v1, or with /v1 but default port 11434)
-    - llama-server endpoints (explicitly configured in llama_server_endpoints)
+    - llama-server / llama-swap endpoints (explicitly configured)
     """
-    cfg = get_config()
-    # Check if it's a llama-server endpoint (has /v1 and is in the configured list)
-    if endpoint in cfg.llama_server_endpoints:
+    # Check if it's a llama-server / llama-swap endpoint (has /v1 and is in a configured list)
+    if is_llama_server(endpoint):
         return False
 
     if "/v1" not in endpoint:
         return False
 
     base_endpoint = endpoint.replace('/v1', '')
-    if base_endpoint in cfg.endpoints:
+    if base_endpoint in get_config().endpoints:
         return False  # It's Ollama's /v1
 
     # Check for default Ollama port
@@ -83,9 +102,9 @@ def is_ext_openai_endpoint(endpoint: str) -> bool:
 def is_openai_compatible(endpoint: str) -> bool:
     """
     Return True if the endpoint speaks the OpenAI API (not native Ollama).
-    This includes external OpenAI endpoints AND llama-server endpoints.
+    This includes external OpenAI endpoints AND llama-server / llama-swap endpoints.
     """
-    return "/v1" in endpoint or endpoint in get_config().llama_server_endpoints
+    return "/v1" in endpoint or is_llama_server(endpoint)
 
 
 def get_tracking_model(endpoint: str, model: str) -> str:
@@ -102,8 +121,8 @@ def get_tracking_model(endpoint: str, model: str) -> str:
     if is_ext_openai_endpoint(endpoint):
         return model
 
-    # llama-server endpoints use normalized names in PS
-    if endpoint in get_config().llama_server_endpoints:
+    # llama-server / llama-swap endpoints use normalized names in PS
+    if is_llama_server(endpoint):
         return _normalize_llama_model_name(model)
 
     # Ollama endpoints: append ":latest" if no version suffix