fix: remove openai provider

Author: YoanSallami

README.md

@@ -1,5 +1,5 @@
 <div align="center">
-<img height=200 src="https://github.com/SynaLinks/synalinks/blob/main/img/synalinks.png?raw=true">
+<img height=200 src="https://github.com/SynaLinks/synalinks/blob/main/img/synalinks.svg?raw=true">
 </div>
 
 <div align="center">
@@ -88,7 +88,7 @@ Building robust LM apps is hard. Synalinks simplifies it with:
 - **Constrained structured outputs** (JSON) for correctness
 - **Automatic async & parallel execution** by default
 - **Metrics, rewards & evaluations** built-in
-- **Native integrations**: OpenAI, Ollama, Anthropic, Mistral, Azure, Groq, Gemini, XAI
+- **Native integrations**: Ollama, Anthropic, Mistral, Azure, Groq, Gemini
 - **Embeddable fast knowledge base support**: based on DuckDB
 - **API-ready**: Deploy with FastAPI or FastMCP
 - **KerasTuner compatibility** for hyperparameter search

coverage-badge.svg

@@ -194,7 +194,5 @@ nav:
       - Synalinks API/Utilities/More plotting utilities.md
       - Synalinks API/Utilities/NLP utilities.md
     - Synalinks API/Config.md
-  - Deployment:
-    - Deployment/Building a REST API.md
   - Observability:
     - Observability/MLflow.md

docs/Deployment/Building a REST API.md

@@ -16,7 +16,7 @@
 
 litellm.drop_params = True
 litellm.disable_aiohttp_transport = True
-
+litellm.drop_params = True
 
 @synalinks_export(
     [
@@ -38,19 +38,6 @@ class LanguageModel(SynalinksSaveable):
 
     For the complete list of models, please refer to the providers documentation.
 
-    **Using OpenAI models**
-
-    ```python
-    import synalinks
-    import os
-
-    os.environ["OPENAI_API_KEY"] = "your-api-key"
-
-    language_model = synalinks.LanguageModel(
-        model="openai/gpt-4o-mini",
-    )
-    ```
-
     **Using Groq models**
 
     ```python
@@ -150,13 +137,13 @@ class LanguageModel(SynalinksSaveable):
     import synalinks
     import os
 
-    os.environ["OPENAI_API_KEY"] = "your-api-key"
+    os.environ["GEMINI_API_KEY"] = "your-api-key"
     os.environ["ANTHROPIC_API_KEY"] = "your-api-key"
 
     language_model = synalinks.LanguageModel(
         model="anthropic/claude-3-sonnet-20240229",
         fallback=synalinks.LanguageModel(
-            model="openai/gpt-4o-mini",
+            model="gemini/gemini-3-flash-preview",
         )
     )
     ```
@@ -228,6 +215,7 @@ async def __call__(self, messages, schema=None, streaming=False, **kwargs):
         json_instance = {}
         input_kwargs = copy.deepcopy(kwargs)
         schema = copy.deepcopy(schema)
+        provider = self.model.split("/")[0]
 
         # Handle reasoning_effort parameter - just forward to litellm if supported
         reasoning_effort = kwargs.pop("reasoning_effort", "none")
@@ -282,7 +270,7 @@ async def __call__(self, messages, schema=None, streaming=False, **kwargs):
                         },
                     }
                 )
-            elif self.model.startswith("openai") or self.model.startswith("azure"):
+            elif self.model.startswith("azure"):
                 # Use constrained structured output for openai
                 # OpenAI require the field  "additionalProperties"
                 # Also OpenAI disallow the field "description" in $ref
@@ -356,8 +344,9 @@ async def __call__(self, messages, schema=None, streaming=False, **kwargs):
         if streaming:
             kwargs.update({"stream": True})
         # Enable prompt caching for the system instructions (that only change during training not inference)
-        system_message_with_cache_control = {**formatted_messages[0], "cache_control": {"type": "ephemeral"}}
-        formatted_messages[0] = system_message_with_cache_control
+        if provider in ("gemini", "anthropic"):
+            system_message_with_cache_control = {**formatted_messages[0], "cache_control": {"type": "ephemeral"}}
+            formatted_messages[0] = system_message_with_cache_control
         for i in range(self.retry):
             try:
                 response_str = ""

img/synalinks.png

@@ -63,7 +63,7 @@ async def test_retrieve_knowledge(self, mock_completion):
         ]
         await knowledge_base.update(docs)
 
-        language_model = LanguageModel(model="openai/gpt-4o-mini")
+        language_model = LanguageModel(model="ollama/mistral")
 
         retrieve_module = RetrieveKnowledge(
             knowledge_base=knowledge_base,
@@ -85,7 +85,7 @@ async def test_retrieve_knowledge_none_input(self):
             data_models=[Document],
         )
 
-        language_model = LanguageModel(model="openai/gpt-4o-mini")
+        language_model = LanguageModel(model="ollama/mistral")
 
         retrieve_module = RetrieveKnowledge(
             knowledge_base=knowledge_base,
@@ -103,7 +103,7 @@ def test_retrieve_knowledge_default_instructions(self):
             data_models=[Document],
         )
 
-        language_model = LanguageModel(model="openai/gpt-4o-mini")
+        language_model = LanguageModel(model="ollama/mistral")
 
         retrieve_module = RetrieveKnowledge(
             knowledge_base=knowledge_base,

img/synalinks.svg

@@ -60,7 +60,7 @@ class Query(DataModel):
         class Answer(DataModel):
             answer: str = Field(description="The correct answer")
 
-        language_model = LanguageModel(model="openai/gpt-4o")
+        language_model = LanguageModel(model="ollama/mistral")
 
         x0 = Input(data_model=Query)
         x1 = await ChainOfThought(