feat(llm): add native groq provider support and fix cache_breakpoint …

lib/crewai/src/crewai/llm.py

@@ -45,6 +45,7 @@
     GEMINI_MODELS,
     OPENAI_MODELS,
 )
+from crewai.llms.cache import CACHE_BREAKPOINT_KEY
 from crewai.utilities import InternalInstructor
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -342,6 +343,7 @@ def _ensure_litellm() -> bool:
     "cerebras",
     "dashscope",
     "snowflake",
+    "groq",
 ]
 
 
@@ -431,6 +433,7 @@ def __new__(cls, model: str, is_litellm: bool = False, **kwargs: Any) -> LLM:
                 "cerebras": "cerebras",
                 "dashscope": "dashscope",
                 "snowflake": "snowflake",
+                "groq": "groq",
             }
 
             canonical_provider = provider_mapping.get(prefix.lower())
@@ -553,6 +556,12 @@ def _matches_provider_pattern(cls, model: str, provider: str) -> bool:
         if provider == "snowflake":
             return True
 
+        if provider == "groq":
+            return any(
+                model_lower.startswith(prefix)
+                for prefix in ["llama", "gemma", "whisper"]
+            )
+
         return False
 
     @classmethod
@@ -664,6 +673,7 @@ def _get_native_provider(cls, provider: str) -> type | None:
             "hosted_vllm",
             "cerebras",
             "dashscope",
+            "groq",
         }
         if provider in openai_compatible_providers:
             from crewai.llms.providers.openai_compatible.completion import (
@@ -2286,6 +2296,16 @@ def _format_messages_for_provider(
                     "Invalid message format. Each message must be a dict with 'role' and 'content' keys"
                 )
 
+        # Strip cache_breakpoint key from messages if not using Anthropic
+        cleaned_messages = []
+        for msg in messages:
+            if not self.is_anthropic:
+                msg_copy = {k: v for k, v in msg.items() if k != CACHE_BREAKPOINT_KEY}
+            else:
+                msg_copy = dict(msg)
+            cleaned_messages.append(msg_copy)
+        messages = cleaned_messages
+
         if "o1" in self.model.lower():
             formatted_messages = []
             for msg in messages:

lib/crewai/src/crewai/llms/providers/openai_compatible/completion.py

@@ -43,6 +43,12 @@ class ProviderConfig:
 
 
 OPENAI_COMPATIBLE_PROVIDERS: dict[str, ProviderConfig] = {
+    "groq": ProviderConfig(
+        base_url="https://api.groq.com/openai/v1",
+        api_key_env="GROQ_API_KEY",
+        base_url_env="GROQ_BASE_URL",
+        api_key_required=True,
+    ),
     "openrouter": ProviderConfig(
         base_url="https://openrouter.ai/api/v1",
         api_key_env="OPENROUTER_API_KEY",

lib/crewai/tests/llms/openai_compatible/test_openai_compatible.py

@@ -95,6 +95,14 @@ def test_dashscope_config(self):
         assert config.api_key_env == "DASHSCOPE_API_KEY"
         assert config.api_key_required is True
 
+    def test_groq_config(self):
+        """Test Groq provider configuration."""
+        config = OPENAI_COMPATIBLE_PROVIDERS["groq"]
+        assert config.base_url == "https://api.groq.com/openai/v1"
+        assert config.api_key_env == "GROQ_API_KEY"
+        assert config.base_url_env == "GROQ_BASE_URL"
+        assert config.api_key_required is True
+
 
 class TestNormalizeOllamaBaseUrl:
     """Tests for _normalize_ollama_base_url helper."""
@@ -271,6 +279,13 @@ def test_llm_creates_openai_compatible_for_dashscope(self):
             assert isinstance(llm, OpenAICompatibleCompletion)
             assert llm.provider == "dashscope"
 
+    def test_llm_creates_openai_compatible_for_groq(self):
+        """Test LLM factory creates OpenAICompatibleCompletion for Groq."""
+        with patch.dict(os.environ, {"GROQ_API_KEY": "test-key"}):
+            llm = LLM(model="groq/llama-3.3-70b-versatile")
+            assert isinstance(llm, OpenAICompatibleCompletion)
+            assert llm.provider == "groq"
+
     def test_llm_with_explicit_provider(self):
         """Test LLM with explicit provider parameter."""
         with patch.dict(os.environ, {"DEEPSEEK_API_KEY": "test-key"}):

lib/crewai/tests/test_llm.py

@@ -806,6 +806,31 @@ def test_ollama_does_not_modify_when_last_is_user(ollama_llm):
     assert formatted == original_messages
 
 
+def test_format_messages_strips_cache_breakpoint_for_non_anthropic():
+    """cache_breakpoint keys must be removed for non-Anthropic providers."""
+    from crewai.llms.cache import CACHE_BREAKPOINT_KEY
+
+    llm = LLM(model="gpt-4o-mini", is_litellm=True)
+    # Simulate a message that carries the cache breakpoint marker
+    messages = [
+        {"role": "user", "content": "Hello", CACHE_BREAKPOINT_KEY: True},
+    ]
+    formatted = llm._format_messages_for_provider(messages)
+    assert all(CACHE_BREAKPOINT_KEY not in msg for msg in formatted)
+
+
+def test_format_messages_preserves_cache_breakpoint_for_anthropic():
+    """cache_breakpoint keys must be preserved for Anthropic providers."""
+    from crewai.llms.cache import CACHE_BREAKPOINT_KEY
+
+    llm = LLM(model="claude-3-5-sonnet-20241022", is_litellm=True)
+    messages = [
+        {"role": "user", "content": "Hello", CACHE_BREAKPOINT_KEY: True},
+    ]
+    formatted = llm._format_messages_for_provider(messages)
+    assert any(CACHE_BREAKPOINT_KEY in msg for msg in formatted)
+
+
 def test_native_provider_raises_error_when_supported_but_fails():
     """Test that when a native provider is in SUPPORTED_NATIVE_PROVIDERS but fails to instantiate, we raise the error."""
     with patch("crewai.llm.SUPPORTED_NATIVE_PROVIDERS", ["openai"]):
@@ -853,6 +878,12 @@ def test_prefixed_models_with_valid_constants_use_native_sdk():
         assert llm3.is_litellm is False
         assert llm3.provider == "gemini"
 
+    # Test groq/ prefix with Groq model in constants → Native SDK
+    with patch.dict(os.environ, {"GROQ_API_KEY": "test-key"}):
+        llm4 = LLM(model="groq/llama-3.1-70b-versatile", is_litellm=False)
+        assert llm4.is_litellm is False
+        assert llm4.provider == "groq"
+
 
 def test_prefixed_models_with_invalid_constants_use_litellm():
     """Test that models with native provider prefixes use LiteLLM when model is NOT in constants and does NOT match patterns."""
@@ -889,10 +920,10 @@ def test_prefixed_models_with_valid_patterns_use_native_sdk():
 
 def test_prefixed_models_with_non_native_providers_use_litellm():
     """Test that models with non-native provider prefixes always use LiteLLM."""
-    # Test groq/ prefix (not a native provider) → LiteLLM
-    llm = LLM(model="groq/llama-3.3-70b", is_litellm=False)
+    # Test replicate/ prefix (not a native provider) → LiteLLM
+    llm = LLM(model="replicate/llama-3-70b", is_litellm=False)
     assert llm.is_litellm is True
-    assert llm.model == "groq/llama-3.3-70b"
+    assert llm.model == "replicate/llama-3-70b"
 
     # Test together/ prefix (not a native provider) → LiteLLM
     llm2 = LLM(model="together/qwen-2.5-72b", is_litellm=False)