Refactor detect_model into shared function, auto-detect context window

Move detect_model() to turnstone.core.model_registry as a single
implementation replacing duplicates in cli.py, server.py, and eval.py.
Auto-detects context window from backend metadata (meta.n_ctx_train)
when available, falling back to the 131072 default otherwise.

Also replace vLLM-specific references in help text and comments with
generic "OpenAI-compatible API" / "model server" language.

turnstone/cli.py

@@ -690,24 +690,11 @@ def _handle_cluster_command(cmd_line: str, console_url: str | None, auth_token:
 # ─── Model auto-detection ─────────────────────────────────────────────────
 
 
-def detect_model(client: OpenAI) -> str:
-    """Auto-detect the model from vLLM's /v1/models endpoint."""
-    try:
-        models = client.models.list()
-        model_ids = [m.id for m in models.data]
-        if not model_ids:
-            print(red("No models found at server. Use --model to specify."))
-            sys.exit(1)
-        if len(model_ids) == 1:
-            return model_ids[0]
-        # Multiple models -- pick first, but inform user
-        print(f"Available models: {', '.join(model_ids)}")
-        print(f"Using: {bold(model_ids[0])} (override with --model)")
-        return model_ids[0]
-    except Exception as e:
-        print(red(f"Could not connect to server: {e}"))
-        print("Is vLLM running? Start it or use --base-url to point elsewhere.")
-        sys.exit(1)
+def detect_model(client: OpenAI) -> tuple[str, int | None]:
+    """Auto-detect model — delegates to :func:`turnstone.core.model_registry.detect_model`."""
+    from turnstone.core.model_registry import detect_model as _detect
+
+    return _detect(client)
 
 
 # ─── Main ──────────────────────────────────────────────────────────────────
@@ -715,7 +702,7 @@ def detect_model(client: OpenAI) -> str:
 
 def main() -> None:
     parser = argparse.ArgumentParser(
-        description="Interactive CLI for vLLM models with tool calling.",
+        description="Interactive CLI for OpenAI-compatible models with tool calling.",
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog=textwrap.dedent("""\
             Examples:
@@ -727,7 +714,7 @@ def main() -> None:
     parser.add_argument(
         "--base-url",
         default="http://localhost:8000/v1",
-        help="vLLM API base URL (default: http://localhost:8000/v1)",
+        help="OpenAI-compatible API base URL (default: http://localhost:8000/v1)",
     )
     parser.add_argument(
         "--model",
@@ -851,7 +838,16 @@ def main() -> None:
         base_url=args.base_url,
         api_key=api_key,
     )
-    model = args.model or detect_model(client)
+    if args.model:
+        model = args.model
+        detected_ctx = None
+    else:
+        model, detected_ctx = detect_model(client)
+
+    # Use detected context window when the user hasn't explicitly set one
+    context_window = args.context_window
+    if detected_ctx and context_window == 131072:  # default unchanged
+        context_window = detected_ctx
 
     # Build model registry (reads [models.*] sections from config.toml)
     from turnstone.core.model_registry import load_model_registry
@@ -860,7 +856,7 @@ def main() -> None:
         base_url=args.base_url,
         api_key=api_key,
         model=model,
-        context_window=args.context_window,
+        context_window=context_window,
     )
 
     # Initialize MCP client (connects to configured MCP servers, if any)

turnstone/core/model_registry.py

@@ -210,3 +210,42 @@ def load_model_registry(
         fallback=fallback,
         agent_model=agent_model,
     )
+
+
+# ---------------------------------------------------------------------------
+# Model auto-detection
+# ---------------------------------------------------------------------------
+
+
+def detect_model(client: OpenAI, log_fn: Any = print) -> tuple[str, int | None]:
+    """Auto-detect the model and context window from the /v1/models endpoint.
+
+    Returns ``(model_id, context_window)`` where *context_window* is
+    ``None`` when the backend does not expose ``meta.n_ctx_train``.
+
+    Calls ``log_fn`` for informational messages (defaults to ``print``).
+    Raises ``SystemExit`` on failure.
+    """
+    try:
+        models = client.models.list()
+        if not models.data:
+            log_fn("Error: No models found at server. Use --model to specify.")
+            raise SystemExit(1)
+        m = models.data[0]
+        if len(models.data) > 1:
+            log_fn(f"Available models: {', '.join(x.id for x in models.data)}")
+            log_fn(f"Using: {m.id} (override with --model)")
+        # Extract context window from backend metadata (llama.cpp, vLLM, etc.)
+        ctx: int | None = None
+        meta = m.model_dump().get("meta")
+        if isinstance(meta, dict):
+            n_ctx = meta.get("n_ctx_train")
+            if isinstance(n_ctx, int) and n_ctx > 0:
+                ctx = n_ctx
+        return m.id, ctx
+    except SystemExit:
+        raise
+    except Exception as e:
+        log_fn(f"Error: Could not connect to server: {e}")
+        log_fn("Is the model server running? Start it or use --base-url to point elsewhere.")
+        raise SystemExit(1) from e

turnstone/core/session.py

@@ -595,7 +595,7 @@ class ChatSession:
         """Stream response, dispatching tokens to the UI as they arrive.
 
         Handles two reasoning delivery mechanisms:
-        1. vLLM's `reasoning_content` field (when --reasoning-parser is set)
+        1. The `reasoning_content` field (e.g. vLLM with --reasoning-parser)
         2. <think>...</think> tags in regular content (common default)
 
         Calls self.ui.on_thinking_stop() on the first received delta.
@@ -715,7 +715,7 @@ class ChatSession:
                 if parts:
                     self.ui.on_info(f"{GRAY}[delta: {', '.join(parts)}]{RESET}")
 
-            # Path 1: reasoning field (vLLM sends as "reasoning" or "reasoning_content")
+            # Path 1: reasoning field (sent as "reasoning" or "reasoning_content")
             rc = getattr(delta, "reasoning", None) or getattr(delta, "reasoning_content", None)
             if rc:
                 _stop_spinner_once()

turnstone/eval.py

@@ -964,7 +964,9 @@ def run_optimization(
     )
 
     if not model:
-        model = _detect_model(client)
+        from turnstone.core.model_registry import detect_model
+
+        model, _ = detect_model(client)
 
     # Load test cases
     with open(test_file) as f:
@@ -1127,18 +1129,6 @@ def run_optimization(
     return results
 
 
-def _detect_model(client: OpenAI) -> str:
-    """Auto-detect the model from the API."""
-    try:
-        models = client.models.list()
-        model_ids = [m.id for m in models.data]
-        if model_ids:
-            return model_ids[0]
-    except Exception:
-        pass
-    raise SystemExit("Could not auto-detect model. Use --model to specify.")
-
-
 # ─── CLI ─────────────────────────────────────────────────────────────────────
 
 

turnstone/server.py

@@ -922,23 +922,11 @@ class ThreadedHTTPServer(ThreadingMixIn, HTTPServer):
 # ---------------------------------------------------------------------------
 
 
-def detect_model(client: OpenAI) -> str:
-    """Auto-detect the model from vLLM's /v1/models endpoint."""
-    try:
-        models = client.models.list()
-        model_ids = [m.id for m in models.data]
-        if not model_ids:
-            print("Error: No models found at server. Use --model to specify.")
-            sys.exit(1)
-        if len(model_ids) == 1:
-            return model_ids[0]
-        print(f"Available models: {', '.join(model_ids)}")
-        print(f"Using: {model_ids[0]} (override with --model)")
-        return model_ids[0]
-    except Exception as e:
-        print(f"Error: Could not connect to server: {e}")
-        print("Is vLLM running? Start it or use --base-url to point elsewhere.")
-        sys.exit(1)
+def detect_model(client: OpenAI) -> tuple[str, int | None]:
+    """Auto-detect model — delegates to :func:`turnstone.core.model_registry.detect_model`."""
+    from turnstone.core.model_registry import detect_model as _detect
+
+    return _detect(client)
 
 
 # ---------------------------------------------------------------------------
@@ -1002,7 +990,7 @@ def main() -> None:
     parser.add_argument(
         "--base-url",
         default="http://localhost:8000/v1",
-        help="vLLM API base URL (default: http://localhost:8000/v1)",
+        help="OpenAI-compatible API base URL (default: http://localhost:8000/v1)",
     )
     parser.add_argument(
         "--model",
@@ -1182,7 +1170,17 @@ def main() -> None:
         base_url=args.base_url,
         api_key=api_key,
     )
-    model = args.model or detect_model(client)
+    if args.model:
+        model = args.model
+        detected_ctx = None
+    else:
+        model, detected_ctx = detect_model(client)
+
+    # Use detected context window when the user hasn't explicitly set one
+    context_window = args.context_window
+    if detected_ctx and context_window == 131072:  # default unchanged
+        context_window = detected_ctx
+        print(f"Context window: {context_window:,} (detected from backend)")
 
     # Build model registry (reads [models.*] sections from config.toml)
     from turnstone.core.model_registry import load_model_registry
@@ -1191,7 +1189,7 @@ def main() -> None:
         base_url=args.base_url,
         api_key=api_key,
         model=model,
-        context_window=args.context_window,
+        context_window=context_window,
     )
 
     # Initialize MCP client (connects to configured MCP servers, if any)