Merge pull request #24 from nuhatech/dev

Release v0.1.23: Introduced `omit_temperature` parameter in `OpenAILL…

Author: mazy06000

CHANGELOG.md

@@ -7,6 +7,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.1.23] - 2026-02-11
+
+### Added
+- **`omit_temperature` parameter for `OpenAILLM`**: Reasoning models (o1, o3, gpt-5-nano, etc.) only support the default temperature value. Set `omit_temperature=True` to omit the `temperature` parameter from API calls entirely.
+- **`omit_temperature` parameter for `AgenticQueryPipeline`**: Propagated to the internally created `OpenAILLM` when no custom LLM is provided.
+
 ## [0.1.22] - 2026-02-11
 
 ### Added
@@ -270,7 +276,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Example scripts for common use cases
 - API reference documentation
 
-[Unreleased]: https://github.com/nuhatech/maktaba/compare/v0.1.22...HEAD
+[Unreleased]: https://github.com/nuhatech/maktaba/compare/v0.1.23...HEAD
+[0.1.23]: https://github.com/nuhatech/maktaba/compare/v0.1.22...v0.1.23
 [0.1.22]: https://github.com/nuhatech/maktaba/compare/v0.1.21...v0.1.22
 [0.1.21]: https://github.com/nuhatech/maktaba/compare/v0.1.20...v0.1.21
 [0.1.20]: https://github.com/nuhatech/maktaba/compare/v0.1.19...v0.1.20

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "maktaba"
-version = "0.1.22"
+version = "0.1.23"
 description = "Production-ready RAG infrastructure for multilingual applications"
 authors = [
     {name = "NuhaTech", email = "contact@nuhatech.com"}

src/maktaba/llm/openai.py

@@ -37,6 +37,7 @@ def __init__(
         timeout_s: float = 30.0,
         prompts: Optional[AgenticPrompts] = None,
         use_max_completion_tokens: bool = False,
+        omit_temperature: bool = False,
     ) -> None:
         """
         Initialize OpenAI LLM.
@@ -50,13 +51,17 @@ def __init__(
             use_max_completion_tokens: Use ``max_completion_tokens`` instead of
                 ``max_tokens`` in API calls. Required for newer OpenAI models
                 (o1, o3, gpt-5-nano, etc.) that no longer accept ``max_tokens``.
+            omit_temperature: Omit the ``temperature`` parameter from API calls.
+                Required for reasoning models (o1, o3, gpt-5-nano, etc.) that
+                only support the default temperature value.
         """
         self.api_key = api_key
         self.model = model
         self.temperature = temperature
         self.timeout_s = timeout_s
         self.prompts = prompts or default_prompts()
         self.use_max_completion_tokens = use_max_completion_tokens
+        self.omit_temperature = omit_temperature
         self._logger = get_logger("maktaba.llm.openai")
 
         # Lazy client initialization
@@ -75,6 +80,16 @@ def _get_client(self) -> Optional[Any]:
             self._client = self._OpenAI(api_key=self.api_key, timeout=self.timeout_s)
         return self._client
 
+    def _temperature_kwargs(self, temperature: float) -> Dict[str, Any]:
+        """Build the temperature keyword argument for the OpenAI API.
+
+        Returns an empty dict when :attr:`omit_temperature` is ``True`` so the
+        parameter is omitted entirely (reasoning models reject custom values).
+        """
+        if self.omit_temperature:
+            return {}
+        return {"temperature": temperature}
+
     def _token_limit_kwargs(self, max_tokens: int | None) -> Dict[str, Any]:
         """Build the token-limit keyword argument for the OpenAI API.
 
@@ -118,7 +133,7 @@ async def complete_text(
                     {"role": "system", "content": system},
                     {"role": "user", "content": prompt},
                 ],
-                temperature=temperature,
+                **self._temperature_kwargs(temperature),
                 **self._token_limit_kwargs(max_tokens),
             )
 
@@ -157,7 +172,7 @@ async def complete_json(
                     {"role": "system", "content": system},
                     {"role": "user", "content": prompt},
                 ],
-                temperature=temperature,
+                **self._temperature_kwargs(temperature),
                 **self._token_limit_kwargs(max_tokens),
                 response_format={"type": "json_object"},
             )
@@ -198,7 +213,7 @@ async def stream_text(
                     {"role": "system", "content": system},
                     {"role": "user", "content": prompt},
                 ],
-                temperature=temperature,
+                **self._temperature_kwargs(temperature),
                 **self._token_limit_kwargs(max_tokens),
                 stream=True,
             )

src/maktaba/pipeline/agentic.py

@@ -44,6 +44,7 @@ def __init__(
         prompts: Optional[AgenticPrompts] = None,
         namespace: Optional[str] = None,
         use_max_completion_tokens: bool = False,
+        omit_temperature: bool = False,
     ) -> None:
         """
         Initialize agentic pipeline.
@@ -61,6 +62,9 @@ def __init__(
             use_max_completion_tokens: Use ``max_completion_tokens`` instead of
                 ``max_tokens`` in OpenAI API calls. Required for newer models
                 (o1, o3, gpt-5-nano, etc.). Only applies when *llm* is not provided.
+            omit_temperature: Omit the ``temperature`` parameter from OpenAI API
+                calls. Required for reasoning models that only support the default
+                value. Only applies when *llm* is not provided.
 
         Example:
             # Use default prompts
@@ -94,6 +98,7 @@ def __init__(
                 model=llm_model,
                 prompts=prompts,
                 use_max_completion_tokens=use_max_completion_tokens,
+                omit_temperature=omit_temperature,
             )
 
     async def _execute_single_query(