diff --git a/app/schemas.py b/app/schemas.py
index e057e48..b0c2c20 100644
--- a/app/schemas.py
+++ b/app/schemas.py
@@ -16,10 +16,10 @@ class ChatMessage(BaseModel):
 class ChatCompletionRequest(BaseModel):
     model: str
     messages: list[ChatMessage] = Field(min_length=1)
-    temperature: float | None = None
-    max_tokens: int | None = None
+    temperature: float | None = Field(default=None, ge=-2.0, le=2.0)
+    max_tokens: int | None = Field(default=None, ge=1)
     top_p: float | None = None
-    n: int | None = None
+    n: int | None = Field(default=None, ge=1)
     stream: bool = False
     stop: str | list[str] | None = None
     presence_penalty: float | None = None
diff --git a/tests/unit/test_chat_request_validation.py b/tests/unit/test_chat_request_validation.py
new file mode 100644
index 0000000..22b2113
--- /dev/null
+++ b/tests/unit/test_chat_request_validation.py
@@ -0,0 +1,71 @@
+"""Reject non-positive max_tokens and n at the schema boundary.
+
+ChatCompletionRequest previously accepted max_tokens=0 and n=0 without
+complaint — both values are meaningless (zero tokens to generate, zero
+choices to return) and most upstream providers reject them with a
+confusing 400. Pydantic now fails closed with a clear validation error
+before the request reaches the routing layer.
+"""
+
+from __future__ import annotations
+
+import pytest
+from pydantic import ValidationError
+
+from app.schemas import ChatCompletionRequest
+
+_BASE = {"model": "gpt-4o", "messages": [{"role": "user", "content": "hi"}]}
+
+
+class TestMaxTokensGuard:
+    def test_zero_rejected(self):
+        with pytest.raises(ValidationError, match="max_tokens"):
+            ChatCompletionRequest.model_validate({**_BASE, "max_tokens": 0})
+
+    def test_negative_rejected(self):
+        with pytest.raises(ValidationError, match="max_tokens"):
+            ChatCompletionRequest.model_validate({**_BASE, "max_tokens": -1})
+
+    def test_positive_accepted(self):
+        req = ChatCompletionRequest.model_validate({**_BASE, "max_tokens": 100})
+        assert req.max_tokens == 100
+
+    def test_none_accepted(self):
+        req = ChatCompletionRequest.model_validate(_BASE)
+        assert req.max_tokens is None
+
+
+class TestNCountGuard:
+    def test_zero_rejected(self):
+        with pytest.raises(ValidationError, match="n"):
+            ChatCompletionRequest.model_validate({**_BASE, "n": 0})
+
+    def test_negative_rejected(self):
+        with pytest.raises(ValidationError, match="n"):
+            ChatCompletionRequest.model_validate({**_BASE, "n": -5})
+
+    def test_positive_accepted(self):
+        req = ChatCompletionRequest.model_validate({**_BASE, "n": 3})
+        assert req.n == 3
+
+    def test_none_accepted(self):
+        req = ChatCompletionRequest.model_validate(_BASE)
+        assert req.n is None
+
+
+class TestTemperatureRange:
+    def test_within_range(self):
+        req = ChatCompletionRequest.model_validate({**_BASE, "temperature": 0.7})
+        assert req.temperature == 0.7
+
+    def test_negative_two_accepted(self):
+        req = ChatCompletionRequest.model_validate({**_BASE, "temperature": -2.0})
+        assert req.temperature == -2.0
+
+    def test_above_two_rejected(self):
+        with pytest.raises(ValidationError, match="temperature"):
+            ChatCompletionRequest.model_validate({**_BASE, "temperature": 2.1})
+
+    def test_below_negative_two_rejected(self):
+        with pytest.raises(ValidationError, match="temperature"):
+            ChatCompletionRequest.model_validate({**_BASE, "temperature": -2.1})