Integrated OpenRouter as model provider

.pre-commit-config.yaml

@@ -16,21 +16,6 @@ repos:
       - id: ruff-format
         name: Run the ruff formatter
 
-  - repo: https://github.com/pre-commit/mirrors-prettier
-    rev: "v4.0.0-alpha.8"
-    hooks:
-      - id: prettier
-        name: Pretty format files
-        exclude: "^daiv/static/"
-        files: "\\.(\
-          css|scss\
-          |js|jsx\
-          |json\
-          |md|markdown|mdown|mkdn\
-          |ts|tsx\
-          |yaml|yml\
-          )$"
-
   - repo: https://github.com/tox-dev/pyproject-fmt
     rev: "v2.5.1"
     hooks:

CHANGELOG.md

@@ -7,6 +7,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## Unreleased
 
+### Added
+
+- Support for `OpenRouter` integration, unified API for LLM providers. **Breaking change: this will be the default provider from now on as it's more reliable and has more models available.**
+
 ## [0.1.0-beta.3] - 2025-03-23
 
 ### Added

daiv/automation/agents/base.py

@@ -10,6 +10,9 @@
 from langchain_core.runnables import Runnable
 from langgraph.graph.state import CompiledStateGraph
 
+from automation.conf import settings
+from core.constants import BOT_NAME
+
 if TYPE_CHECKING:
     from langchain_core.language_models.chat_models import BaseChatModel
     from langchain_core.messages import BaseMessage
@@ -22,6 +25,7 @@ class ModelProvider(StrEnum):
     ANTHROPIC = "anthropic"
     OPENAI = "openai"
     GOOGLE_GENAI = "google_genai"
+    OPENROUTER = "openrouter"
 
 
 class ThinkingLevel(StrEnum):
@@ -78,10 +82,14 @@ def get_model(self, *, model: str, thinking_level: ThinkingLevel | None = None,
         Returns:
             BaseChatModel: The model instance
         """
-        model_kwargs = self.get_model_kwargs(model=model, thinking_level=thinking_level, **kwargs)
+        model_kwargs = self.get_model_kwargs(
+            model=model, model_provider=BaseAgent.get_model_provider(model), thinking_level=thinking_level, **kwargs
+        )
         return init_chat_model(**model_kwargs)
 
-    def get_model_kwargs(self, *, thinking_level: ThinkingLevel | None = None, **kwargs) -> dict:
+    def get_model_kwargs(
+        self, *, model_provider: ModelProvider, thinking_level: ThinkingLevel | None = None, **kwargs
+    ) -> dict:
         """
         Get the keyword arguments to pass to the model.
 
@@ -93,28 +101,21 @@ def get_model_kwargs(self, *, thinking_level: ThinkingLevel | None = None, **kwa
             "callbacks": [self.usage_handler],
             "configurable_fields": ("temperature", "max_tokens"),
             "model_kwargs": {},
+            "model_provider": model_provider,
             **kwargs,
         }
 
-        model_provider = BaseAgent.get_model_provider(_kwargs["model"])
-
         if model_provider == ModelProvider.ANTHROPIC:
-            # As stated in docs: https://docs.anthropic.com/en/api/rate-limits#updated-rate-limits
-            # the OTPM is calculated based on the max_tokens. We need to use a fair value to avoid rate limiting.
-            # If needed, we can increase this value using the configurable field.
             if thinking_level and _kwargs["model"].startswith("claude-3-7-sonnet"):
+                max_tokens, thinking_tokens = self._get_anthropic_thinking_tokens(thinking_level=thinking_level)
                 # When using thinking the temperature need to be set to 1
                 _kwargs["temperature"] = 1
-                if thinking_level == ThinkingLevel.LOW:
-                    _kwargs["max_tokens"] = 4_096
-                    _kwargs["thinking"] = {"type": "enabled", "budget_tokens": 2_048}
-                elif thinking_level == ThinkingLevel.MEDIUM:
-                    _kwargs["max_tokens"] = 10_240
-                    _kwargs["thinking"] = {"type": "enabled", "budget_tokens": 7_168}
-                elif thinking_level == ThinkingLevel.HIGH:
-                    _kwargs["max_tokens"] = 20_480
-                    _kwargs["thinking"] = {"type": "enabled", "budget_tokens": 16_384}
+                _kwargs["max_tokens"] = max_tokens
+                _kwargs["thinking"] = {"type": "enabled", "budget_tokens": thinking_tokens}
             else:
+                # As stated in docs: https://docs.anthropic.com/en/api/rate-limits#updated-rate-limits
+                # the OTPM is calculated based on the max_tokens. We need to use a fair value to avoid rate limiting.
+                # If needed, we can increase this value using the configurable field.
                 _kwargs["max_tokens"] = 2_048
 
             if _kwargs["model"].startswith("claude-3-7-sonnet"):
@@ -125,11 +126,42 @@ def get_model_kwargs(self, *, thinking_level: ThinkingLevel | None = None, **kwa
             if thinking_level and _kwargs["model"].startswith(("o1", "o3")):
                 _kwargs["temperature"] = 1
                 _kwargs["reasoning_effort"] = thinking_level
-        elif model_provider in [ModelProvider.GOOGLE_GENAI]:
-            # otherwise google_genai will be inferred as google_vertexai
-            _kwargs["model_provider"] = model_provider
+
+        elif model_provider == ModelProvider.OPENROUTER:
+            _kwargs["model"] = _kwargs["model"].split(":")[1]
+            # OpenRouter is OpenAI compatible, so we need to use the OpenAI model provider
+            _kwargs["model_provider"] = ModelProvider.OPENAI
+            _kwargs["model_kwargs"]["extra_headers"] = {
+                "HTTP-Referer": "https://github.com/srtab/daiv",
+                "X-Title": BOT_NAME,
+            }
+            _kwargs["openai_api_base"] = settings.OPENROUTER_API_BASE
+            _kwargs["openai_api_key"] = settings.OPENROUTER_API_KEY
+
+            if _kwargs["model"].startswith("anthropic/claude-3-7-sonnet"):
+                _kwargs["model_kwargs"]["extra_headers"]["anthropic-beta"] = "token-efficient-tools-2025-02-19"
+
+            if thinking_level:
+                _kwargs["temperature"] = 1
+                _kwargs["extra_body"] = {"reasoning": {"effort": thinking_level}}
+
+            elif _kwargs["model"].startswith("anthropic"):
+                # Avoid rate limiting by setting a fair max_tokens value
+                _kwargs["max_tokens"] = 2_048
+
         return _kwargs
 
+    def _get_anthropic_thinking_tokens(self, *, thinking_level: ThinkingLevel) -> tuple[int, dict]:
+        """
+        Get the thinking tokens and max tokens for the model.
+        """
+        if thinking_level == ThinkingLevel.LOW:
+            return 4_096, {"type": "enabled", "budget_tokens": 2_048}
+        elif thinking_level == ThinkingLevel.MEDIUM:
+            return 32_768, {"type": "enabled", "budget_tokens": 25_600}
+        elif thinking_level == ThinkingLevel.HIGH:
+            return 64_000, {"type": "enabled", "budget_tokens": 55_808}
+
     def draw_mermaid(self):
         """
         Draw the graph in Mermaid format.
@@ -194,7 +226,10 @@ def get_model_provider(model_name: str) -> ModelProvider:
         model_provider: ModelProvider | None = None
 
         if model_name.startswith("gemini"):
+            # The _attempt_infer_model_provider will return google_vertexai instead of google_genai
             model_provider = ModelProvider.GOOGLE_GENAI
+        elif model_name.startswith("openrouter:"):
+            model_provider = ModelProvider.OPENROUTER
         else:
             model_provider = cast("ModelProvider | None", _attempt_infer_model_provider(model_name))
 

daiv/automation/agents/codebase_chat/conf.py

@@ -13,7 +13,7 @@ class CodebaseChatSettings(BaseSettings):
 
     NAME: str = Field(default="CodebaseChat", description="Name of the codebase chat agent.")
     MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022, description="Model name to be used for codebase chat."
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Model name to be used for codebase chat."
     )
     TEMPERATURE: float = Field(default=0.5, description="Temperature to be used for codebase chat.")
 

daiv/automation/agents/codebase_search/conf.py

@@ -14,17 +14,16 @@ class CodebaseSearchSettings(BaseSettings):
     NAME: str = Field(default="CodebaseSearch", description="Name of the codebase search agent.")
     TOP_N: int = Field(default=10, description="Number of results to return from the codebase search.")
     REPHRASE_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Model name to be used for codebase search."
+        default=ModelName.GPT_4O_MINI, description="Model name to be used for codebase search."
     )
     REPHRASE_FALLBACK_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022, description="Fallback model name to be used for codebase search."
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Fallback model name to be used for codebase search."
     )
     RERANKING_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Model name to be used for listwise reranking."
+        default=ModelName.GPT_4O_MINI, description="Model name to be used for listwise reranking."
     )
     RERANKING_FALLBACK_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022,
-        description="Fallback model name to be used for listwise reranking.",
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Fallback model name to be used for listwise reranking."
     )
 
 

daiv/automation/agents/constants.py

@@ -2,12 +2,18 @@
 
 
 class ModelName(StrEnum):
-    CLAUDE_3_7_SONNET_20250219 = "claude-3-7-sonnet-20250219"
-    CLAUDE_3_5_HAIKU_20241022 = "claude-3-5-haiku-20241022"
-    GPT_4O_2024_11_20 = "gpt-4o-2024-11-20"
-    GPT_4O_MINI_2024_07_18 = "gpt-4o-mini-2024-07-18"
-    O1_2024_12_17 = "o1-2024-12-17"
-    O3_MINI_2025_01_31 = "o3-mini-2025-01-31"
-    GEMINI_2_0_FLASH = "gemini-2.0-flash"
-    GEMINI_2_0_FLASH_LITE = "gemini-2.0-flash-lite"
-    GEMINI_2_0_PRO = "gemini-2.0-pro-exp-02-05"
+    """
+    `openrouter` provider is the default provider to use any model that is supported by OpenRouter.
+
+    You can also use `anthropic`, `google` or `openai` model providers directly to use any model that is supported
+    by Anthropic, Google or OpenAI.
+    """
+
+    CLAUDE_3_7_SONNET = "openrouter:anthropic/claude-3-7-sonnet"
+    CLAUDE_3_5_HAIKU = "openrouter:anthropic/claude-3-5-haiku"
+    GPT_4O = "openrouter:openai/gpt-4o"
+    GPT_4O_MINI = "openrouter:openai/gpt-4o-mini"
+    O1 = "openrouter:openai/o1"
+    O3_MINI = "openrouter:openai/o3-mini"
+    GEMINI_2_0_FLASH = "openrouter:google/gemini-2.0-flash-001"
+    GEMINI_2_0_FLASH_LITE = "openrouter:google/gemini-2.0-flash-lite-001"

daiv/automation/agents/image_url_extractor/conf.py

@@ -13,7 +13,7 @@ class ImageURLExtractorSettings(BaseSettings):
 
     NAME: str = Field(default="ImageURLExtractor", description="Name of the image URL extractor agent.")
     MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Model name to be used for image URL extractor."
+        default=ModelName.GPT_4O_MINI, description="Model name to be used for image URL extractor."
     )
 
 

daiv/automation/agents/issue_addressor/conf.py

@@ -14,10 +14,10 @@ class IssueAddressorSettings(BaseSettings):
     NAME: str = Field(default="IssueAddressor", description="Name of the issue addressor agent.")
     RECURSION_LIMIT: int = Field(default=50, description="Recursion limit for the issue addressor agent.")
     ASSESSMENT_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Model name to be used for issue assessment."
+        default=ModelName.GPT_4O_MINI, description="Model name to be used for issue assessment."
     )
     FALLBACK_ASSESSMENT_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022, description="Fallback model name to be used for issue assessment."
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Fallback model name to be used for issue assessment."
     )
 
 

daiv/automation/agents/pipeline_fixer/conf.py

@@ -15,22 +15,22 @@ class PipelineFixerSettings(BaseSettings):
     NAME: str = Field(default="PipelineFixer", description="Name of the pipeline fixer agent.")
     MAX_ITERATIONS: int = Field(default=20, description="Maximum number of retry iterations for pipeline fixer.")
     LOG_EVALUATOR_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022, description="Model name to be used for log evaluator."
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Model name to be used for log evaluator."
     )
     LOG_EVALUATOR_FALLBACK_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_7_SONNET_20250219, description="Fallback model name for log evaluator."
+        default=ModelName.CLAUDE_3_7_SONNET, description="Fallback model name for log evaluator."
     )
     TROUBLESHOOTING_MODEL_NAME: ModelName = Field(
-        default=ModelName.O3_MINI_2025_01_31, description="Model name to be used for pipeline fixer."
+        default=ModelName.O3_MINI, description="Model name to be used for pipeline fixer."
     )
     TROUBLESHOOTING_THINKING_LEVEL: ThinkingLevel = Field(
         default=ThinkingLevel.HIGH, description="Thinking level to be used for pipeline fixer."
     )
     LINT_EVALUATOR_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022, description="Model name to be used for lint evaluator."
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Model name to be used for lint evaluator."
     )
     LINT_EVALUATOR_FALLBACK_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Fallback model name for lint evaluator."
+        default=ModelName.GPT_4O_MINI, description="Fallback model name for lint evaluator."
     )
 
 

daiv/automation/agents/plan_and_execute/conf.py

@@ -14,19 +14,19 @@ class PlanAndExecuteSettings(BaseSettings):
 
     NAME: str = Field(default="PlanAndExecute", description="Name of the plan and execute agent.")
     PLANNING_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_7_SONNET_20250219, description="Model name to be used to plan tasks."
+        default=ModelName.CLAUDE_3_7_SONNET, description="Model name to be used to plan tasks."
     )
     PLANNING_THINKING_LEVEL: ThinkingLevel = Field(
         default=ThinkingLevel.MEDIUM, description="Thinking level to be used to plan tasks."
     )
     EXECUTION_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_7_SONNET_20250219, description="Model name to be used to execute tasks."
+        default=ModelName.CLAUDE_3_7_SONNET, description="Model name to be used to execute tasks."
     )
     EXECUTION_THINKING_LEVEL: ThinkingLevel = Field(
         default=ThinkingLevel.MEDIUM, description="Thinking level to be used to execute tasks."
     )
     PLAN_APPROVAL_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Model name to be used to evaluate the plan approval."
+        default=ModelName.GPT_4O_MINI, description="Model name to be used to evaluate the plan approval."
     )
 
 

daiv/automation/agents/pr_describer/conf.py

@@ -9,10 +9,10 @@ class PRDescriberSettings(BaseSettings):
 
     NAME: str = Field(default="PullRequestDescriber", description="Name of the PR describer agent.")
     MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022, description="Model name to be used for PR describer."
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Model name to be used for PR describer."
     )
     FALLBACK_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Fallback model name to be used for PR describer."
+        default=ModelName.GPT_4O_MINI, description="Fallback model name to be used for PR describer."
     )
 
 

daiv/automation/agents/review_addressor/conf.py

@@ -9,17 +9,16 @@ class ReviewAddressorSettings(BaseSettings):
 
     NAME: str = Field(default="ReviewAddressor", description="Name of the review addressor agent.")
     ASSESSMENT_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022, description="Model name to be used for review assessment."
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Model name to be used for review assessment."
     )
     FALLBACK_ASSESSMENT_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Fallback model name to be used for review assessment."
+        default=ModelName.GPT_4O_MINI, description="Fallback model name to be used for review assessment."
     )
     REPLY_MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022,
-        description="Model name to be used for reply to comments or questions.",
+        default=ModelName.CLAUDE_3_5_HAIKU, description="Model name to be used for reply to comments or questions."
     )
     FALLBACK_REPLY_MODEL_NAME: ModelName = Field(
-        default=ModelName.GPT_4O_MINI_2024_07_18, description="Fallback model name for REPLY_MODEL_NAME."
+        default=ModelName.GPT_4O_MINI, description="Fallback model name for REPLY_MODEL_NAME."
     )
     REPLY_TEMPERATURE: float = Field(default=0.5, description="Temperature for the reply model.")
 

daiv/automation/agents/snippet_replacer/conf.py

@@ -11,7 +11,7 @@ class SnippetReplacerSettings(BaseSettings):
 
     NAME: str = Field(default="SnippetReplacer", description="Name of the snippet replacer agent.")
     MODEL_NAME: ModelName = Field(
-        default=ModelName.CLAUDE_3_5_HAIKU_20241022,
+        default=ModelName.CLAUDE_3_5_HAIKU,
         description="Model name to be used for snippet replacer. This model is used for the LLM strategy.",
     )
     STRATEGY: Literal["llm", "find_and_replace"] = Field(

daiv/automation/conf.py

@@ -7,6 +7,12 @@
 class AutomationSettings(BaseSettings):
     model_config = SettingsConfigDict(secrets_dir="/run/secrets", env_prefix="AUTOMATION_")
 
+    # OpenRouter settings
+    OPENROUTER_API_KEY: str | None = Field(default=None, description="OpenRouter API key", alias="OPENROUTER_API_KEY")
+    OPENROUTER_API_BASE: str | None = Field(
+        default="https://openrouter.ai/api/v1", description="OpenRouter API base url", alias="OPENROUTER_API_BASE"
+    )
+
     # Web search settings
     WEB_SEARCH_MAX_RESULTS: int = Field(default=5, description="Maximum number of results to return from web search")
     WEB_SEARCH_ENGINE: Literal["duckduckgo", "tavily"] = Field(

docker/local/app/config.secrets.example.env

@@ -5,6 +5,7 @@ CODEBASE_GITLAB_AUTH_TOKEN=
 OPENAI_API_KEY=
 ANTHROPIC_API_KEY=
 GOOGLE_API_KEY=
+OPENROUTER_API_KEY=
 
 # Monitoring
 LANGCHAIN_API_KEY=