diff --git a/docs/dsh-architecture.md b/docs/dsh-architecture.md
index d7c2f1b..7d045c8 100644
--- a/docs/dsh-architecture.md
+++ b/docs/dsh-architecture.md
@@ -88,6 +88,8 @@ is added to `MODEL_CHOICES` **and** `MODEL_LABELS`.
 
 A chosen model is validated server-side (`if active_model not in MODEL_CHOICES:
 active_model = LLM_MODEL`), so an unlisted id silently falls back to the default.
+That also means removing an id is safe: sessions that still reference it fall back
+rather than erroring.
 
 ### Adding a model
 
@@ -100,13 +102,15 @@ active_model = LLM_MODEL`), so an unlisted id silently falls back to the default
    bakes `app.py` via `COPY . .`, so a rebuild is required — restarting is not enough).
 4. Verify: `curl -s http://127.0.0.1:3081/api/models` should list the new id.
 
-### Local Ollama models
+### The default model is not necessarily a working model
 
-The `.27` models are exposed by OmniRoute as `ollama/qwen2.5:14b` and
-`ollama/qwen2.5-coder:14b` (`ollama-local/…` also exists but is duplicated).
-Because Ollama now starts **off** at boot, these fail until it is started from the
-family console; hence the `needs Ollama on` label. Inference is CPU-only on `.27`,
-so expect it to be slower than the cloud routes.
+`DSH_LLM_MODEL` (set to `auto/best-chat` in `docker-compose.yml`) is the default.
+The `auto/*` entries are **OmniRoute routers**, not providers: if OmniRoute's
+candidate pool is exhausted it returns `service_unavailable` / "Maximum combo
+retry limit reached", and DSH surfaces that as *"Model returned no response"*.
+That is an OmniRoute-side failure, not a DSH bug — check it by calling OmniRoute
+directly, and prefer an explicit model id (e.g. `deepseek/deepseek-v4-flash`)
+when testing whether DSH itself is healthy.
 
 ---
 
diff --git a/dsh/app.py b/dsh/app.py
index 55c9637..b2c3b56 100644
--- a/dsh/app.py
+++ b/dsh/app.py
@@ -665,11 +665,6 @@ MODEL_CHOICES = ["auto/best-chat",
                  "auto/best-reasoning",
                  "auto/best-coding",
                  "auto/best-vision",
-                 # Local models on .27, reached through OmniRoute. Free to run, but
-                 # Ollama must be started first (family console -> AI Tools ->
-                 # Ollama Local LLM). CPU inference, so slower than the cloud ones.
-                 "ollama/qwen2.5:14b",
-                 "ollama/qwen2.5-coder:14b",
                  "openrouter/deepseek/deepseek-v4-flash",
                  "openrouter/deepseek/deepseek-v4-pro"]
 MODEL_LABELS = {
@@ -681,8 +676,6 @@ MODEL_LABELS = {
     "auto/best-reasoning": "Reasoning · OmniRoute best",
     "auto/best-coding": "Coding · OmniRoute best",
     "auto/best-vision": "Multimodal · OmniRoute vision",
-    "ollama/qwen2.5:14b": "Local · Qwen2.5 14B (.27, needs Ollama on)",
-    "ollama/qwen2.5-coder:14b": "Local · Qwen2.5 Coder 14B (.27, needs Ollama on)",
     "openrouter/deepseek/deepseek-v4-flash": "DeepSeek v4 flash (OpenRouter)",
     "openrouter/deepseek/deepseek-v4-pro": "DeepSeek v4 pro (OpenRouter)",
 }