diff --git a/src/forge/clients/llamafile.py b/src/forge/clients/llamafile.py
index f8082cc..481f0bd 100644
--- a/src/forge/clients/llamafile.py
+++ b/src/forge/clients/llamafile.py
@@ -142,6 +142,7 @@ def __init__(
         cache_prompt: bool = True,
         slot_id: int | None = None,
         recommended_sampling: bool = False,
+        api_key: str | None = None,
     ) -> None:
         self.base_url = base_url
         # gguf_path is the canonical identity. self.model is the stem (no
@@ -149,7 +150,10 @@ def __init__(
         # (llama-server ignores it but it flows into eval JSONL rows) and
         # for sampling-defaults lookup.
         self.gguf_path = Path(gguf_path)
-        self.model = self.gguf_path.stem
+        # When gguf_path is a real file, use stem (no .gguf/.llamafile suffix).
+        # When it's a plain model name (no GGUF extension), use it as-is.
+        _GGUF_SUFFIXES = {".gguf", ".llamafile"}
+        self.model = self.gguf_path.stem if self.gguf_path.suffix.lower() in _GGUF_SUFFIXES else str(gguf_path)
         # Apply per-model recommended sampling defaults. Caller's explicit
         # (non-None) kwargs win over the map field-by-field.
         defaults = apply_sampling_defaults(self.model, strict=recommended_sampling)
@@ -160,7 +164,8 @@ def __init__(
         self.repeat_penalty = repeat_penalty if repeat_penalty is not None else defaults.get("repeat_penalty")
         self.presence_penalty = presence_penalty if presence_penalty is not None else defaults.get("presence_penalty")
         self.mode = mode
-        self._http = httpx.AsyncClient(timeout=timeout)
+        headers = {"Authorization": f"Bearer {api_key}"} if api_key else {}
+        self._http = httpx.AsyncClient(timeout=timeout, headers=headers)
         self._think: bool = think if think is not None else True  # auto = capture
         self._cache_prompt = cache_prompt
         self._slot_id = slot_id
@@ -407,6 +412,8 @@ async def get_context_length(self) -> int | None:
             base = base[:-3]
 
         resp = await self._http.get(f"{base}/props")
+        if resp.status_code == 404:
+            return None
         resp.raise_for_status()
         data = resp.json()
 
diff --git a/src/forge/proxy/__main__.py b/src/forge/proxy/__main__.py
index efb4c65..774a4d2 100644
--- a/src/forge/proxy/__main__.py
+++ b/src/forge/proxy/__main__.py
@@ -50,6 +50,7 @@ def main() -> None:
     parser.add_argument("--max-retries", type=int, default=3, help="Max retries per request (default: 3)")
     parser.add_argument("--no-rescue", action="store_true", help="Disable rescue parsing")
     parser.add_argument("--verbose", "-v", action="store_true", help="Verbose logging")
+    parser.add_argument("--api-key", help="API key for backend authentication (sent as Bearer token)")
 
     args = parser.parse_args()
 
@@ -82,6 +83,7 @@ def main() -> None:
         serialize=serialize,
         max_retries=args.max_retries,
         rescue_enabled=not args.no_rescue,
+        api_key=args.api_key,
     )
 
     def _shutdown(sig: int, _frame: object) -> None:
diff --git a/src/forge/proxy/proxy.py b/src/forge/proxy/proxy.py
index 330d138..f99b846 100644
--- a/src/forge/proxy/proxy.py
+++ b/src/forge/proxy/proxy.py
@@ -59,6 +59,7 @@ def __init__(
         serialize: bool | None = None,
         max_retries: int = 3,
         rescue_enabled: bool = True,
+        api_key: str | None = None,
     ) -> None:
         """
         Args:
@@ -92,6 +93,7 @@ def __init__(
         self._port = port
         self._max_retries = max_retries
         self._rescue_enabled = rescue_enabled
+        self._api_key = api_key
 
         # Auto-detect serialization: managed = single GPU = serialize
         if serialize is None:
@@ -164,12 +166,13 @@ async def _async_start(self, ready: threading.Event) -> None:
             if not base.endswith("/v1"):
                 base = base + "/v1"
             # External mode: caller manages the backend, so we don't have a
-            # GGUF path. "default" is a placeholder identity for the wire
-            # model field (llama-server ignores it) and JSONL model field.
+            # GGUF path. Use model name if provided, else "default" placeholder.
+            wire_model = self._model or "default"
             client = LlamafileClient(
-                gguf_path="default",
+                gguf_path=wire_model,
                 base_url=base,
                 mode="native",
+                api_key=self._api_key,
             )
             if self._budget_tokens is not None:
                 budget = self._budget_tokens