diff --git a/pyproject.toml b/pyproject.toml
index 660fc11f..aa0c0cf0 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -93,6 +93,11 @@ asyncio_default_fixture_loop_scope = "session"
 addopts = "--strict-markers --cov=src/ --cov=sdks/python/src/honcho --cov-report=term-missing --ignore=tests/alembic"
 testpaths = ["tests"]
 pythonpath = ["src"]
+filterwarnings = [
+    "ignore::DeprecationWarning:cashews.*:",
+    "ignore::DeprecationWarning:websockets.*:",
+    "ignore::DeprecationWarning:uvicorn.*:",
+]
 
 [tool.coverage.report]
 exclude_lines = [
diff --git a/src/config.py b/src/config.py
index f01c451b..a500df5a 100644
--- a/src/config.py
+++ b/src/config.py
@@ -251,6 +251,9 @@ class DeriverSettings(BackupLLMSettingsMixin, HonchoSettings):
     PROVIDER: SupportedProviders = "google"
     MODEL: str = "gemini-2.5-flash-lite"
     TEMPERATURE: float | None = None
+    TOP_P: float | None = None
+    REPETITION_PENALTY: float | None = None
+    NO_REPEAT_NGRAM_SIZE: int | None = None
 
     # Whether to deduplicate documents when creating them
     DEDUPLICATE: bool = True
diff --git a/src/deriver/deriver.py b/src/deriver/deriver.py
index b8735e3c..022e0d77 100644
--- a/src/deriver/deriver.py
+++ b/src/deriver/deriver.py
@@ -131,6 +131,9 @@ async def process_representation_tasks_batch(
         response_model=PromptRepresentation,
         json_mode=True,
         temperature=settings.DERIVER.TEMPERATURE,
+        top_p=settings.DERIVER.TOP_P,
+        repetition_penalty=settings.DERIVER.REPETITION_PENALTY,
+        no_repeat_ngram_size=settings.DERIVER.NO_REPEAT_NGRAM_SIZE,
         stop_seqs=["   \n", "\n\n\n\n"],
         thinking_budget_tokens=settings.DERIVER.THINKING_BUDGET_TOKENS,
         max_input_tokens=settings.DERIVER.MAX_INPUT_TOKENS,
diff --git a/src/utils/clients.py b/src/utils/clients.py
index ae1beadd..39dac35d 100644
--- a/src/utils/clients.py
+++ b/src/utils/clients.py
@@ -573,6 +573,9 @@ async def _stream_final_response(
     response_model: type[BaseModel] | None,
     json_mode: bool,
     temperature: float | None,
+    top_p: float | None,
+    repetition_penalty: float | None,
+    no_repeat_ngram_size: int | None,
     stop_seqs: list[str] | None,
     reasoning_effort: ReasoningEffortType,
     verbosity: VerbosityType,
@@ -616,6 +619,9 @@ async def _stream_final_response(
         response_model,
         json_mode,
         _get_effective_temperature(temperature),
+        top_p,
+        repetition_penalty,
+        no_repeat_ngram_size,
         stop_seqs,
         reasoning_effort,
         verbosity,
@@ -643,6 +649,9 @@ async def _execute_tool_loop(
     response_model: type[BaseModel] | None,
     json_mode: bool,
     temperature: float | None,
+    top_p: float | None,
+    repetition_penalty: float | None,
+    no_repeat_ngram_size: int | None,
     stop_seqs: list[str] | None,
     reasoning_effort: ReasoningEffortType,
     verbosity: VerbosityType,
@@ -745,6 +754,9 @@ async def _execute_tool_loop(
                 response_model,
                 json_mode,
                 _get_effective_temperature(temperature),
+                top_p,
+                repetition_penalty,
+                no_repeat_ngram_size,
                 stop_seqs,
                 gpt5_reasoning_effort,
                 gpt5_verbosity,
@@ -808,6 +820,9 @@ async def _execute_tool_loop(
                     response_model=response_model,
                     json_mode=json_mode,
                     temperature=temperature,
+                    top_p=top_p,
+                    repetition_penalty=repetition_penalty,
+                    no_repeat_ngram_size=no_repeat_ngram_size,
                     stop_seqs=stop_seqs,
                     reasoning_effort=reasoning_effort,
                     verbosity=verbosity,
@@ -941,6 +956,9 @@ async def _execute_tool_loop(
             response_model=response_model,
             json_mode=json_mode,
             temperature=temperature,
+            top_p=top_p,
+            repetition_penalty=repetition_penalty,
+            no_repeat_ngram_size=no_repeat_ngram_size,
             stop_seqs=stop_seqs,
             reasoning_effort=reasoning_effort,
             verbosity=verbosity,
@@ -977,6 +995,9 @@ async def _execute_tool_loop(
             response_model,
             json_mode,
             _get_effective_temperature(temperature),
+            top_p,
+            repetition_penalty,
+            no_repeat_ngram_size,
             stop_seqs,
             reasoning_effort,
             verbosity,
@@ -1182,6 +1203,9 @@ async def honcho_llm_call(
     response_model: type[M],
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1211,6 +1235,9 @@ async def honcho_llm_call(
     response_model: None = None,
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1240,6 +1267,9 @@ async def honcho_llm_call(
     response_model: type[BaseModel] | None = None,
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1269,6 +1299,9 @@ async def honcho_llm_call(
     response_model: type[BaseModel] | None = None,
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1422,6 +1455,9 @@ async def honcho_llm_call(
                 response_model,
                 json_mode,
                 _get_effective_temperature(temperature),
+                top_p,
+                repetition_penalty,
+                no_repeat_ngram_size,
                 stop_seqs,
                 gpt5_reasoning_effort,
                 gpt5_verbosity,
@@ -1439,6 +1475,9 @@ async def honcho_llm_call(
                 response_model,
                 json_mode,
                 _get_effective_temperature(temperature),
+                top_p,
+                repetition_penalty,
+                no_repeat_ngram_size,
                 stop_seqs,
                 gpt5_reasoning_effort,
                 gpt5_verbosity,
@@ -1522,6 +1561,9 @@ async def honcho_llm_call(
         response_model=response_model,
         json_mode=json_mode,
         temperature=temperature,
+        top_p=top_p,
+        repetition_penalty=repetition_penalty,
+        no_repeat_ngram_size=no_repeat_ngram_size,
         stop_seqs=stop_seqs,
         reasoning_effort=reasoning_effort,
         verbosity=verbosity,
@@ -1559,6 +1601,9 @@ async def honcho_llm_call_inner(
     response_model: type[M],
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1580,6 +1625,9 @@ async def honcho_llm_call_inner(
     response_model: None = None,
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1601,6 +1649,9 @@ async def honcho_llm_call_inner(
     response_model: type[BaseModel] | None = None,
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1621,6 +1672,9 @@ async def honcho_llm_call_inner(
     response_model: type[BaseModel] | None = None,
     json_mode: bool = False,
     temperature: float | None = None,
+    top_p: float | None = None,
+    repetition_penalty: float | None = None,
+    no_repeat_ngram_size: int | None = None,
     stop_seqs: list[str] | None = None,
     reasoning_effort: Literal["low", "medium", "high", "minimal"]
     | None = None,  # OpenAI only
@@ -1865,6 +1919,21 @@ async def honcho_llm_call_inner(
             if temperature is not None and "gpt-5" not in model:
                 openai_params["temperature"] = temperature
 
+            # Add generation parameters for explicit control
+            if top_p is not None:
+                openai_params["top_p"] = top_p
+
+            # Add vLLM-specific parameters only for vLLM provider
+            # These are not standard OpenAI API parameters and must be passed via extra_body
+            if provider == "vllm":
+                extra_body: dict[str, Any] = {}
+                if repetition_penalty is not None:
+                    extra_body["repetition_penalty"] = repetition_penalty
+                if no_repeat_ngram_size is not None:
+                    extra_body["no_repeat_ngram_size"] = no_repeat_ngram_size
+                if extra_body:
+                    openai_params["extra_body"] = extra_body
+
             if "gpt-5" in model:
                 openai_params["max_completion_tokens"] = params["max_tokens"]
                 if reasoning_effort: