feat(realism/llm): DB-backed LLMConfig, factory DB-first dispatch, Ollama HTTP mode

decnet/realism/llm/factory.py

@@ -24,7 +24,14 @@ from decnet.realism.llm.base import LLMBackend
 
 
 def get_llm(*, model: str | None = None, **kwargs: Any) -> LLMBackend:
-    """Instantiate the LLM backend selected by environment.
+    """Instantiate the LLM backend selected by DB config or environment.
+
+    Resolution order:
+    1. Process-level cached backend (populated by the DB config row via
+       :func:`decnet.realism.llm.config.apply`).  Returned as-is when
+       *model* and *kwargs* are both absent — the common case.
+    2. Env-var path (``DECNET_REALISM_LLM`` / ``DECNET_REALISM_MODEL`` /
+       ``DECNET_REALISM_TIMEOUT``) — legacy / default-install fallback.
 
     *model* (when provided) overrides whatever the backend's own default
     is — e.g. for :class:`OllamaBackend` that's ``llama3.1`` unless
@@ -32,6 +39,13 @@ def get_llm(*, model: str | None = None, **kwargs: Any) -> LLMBackend:
     ``decnet orchestrate --model gpt-oss`` without each backend having
     to know about CLI flags.
     """
+    # Fast path: DB-configured cached backend.
+    if model is None and not kwargs:
+        from decnet.realism.llm.config import get_cached_backend
+        cached = get_cached_backend()
+        if cached is not None:
+            return cached
+
     backend_key = os.environ.get("DECNET_REALISM_LLM", "ollama").lower()
 
     if backend_key == "ollama":