feat: add multi-model support for per-analysis LLM selection

Allow users to choose the LLM model on a per-analysis basis. The
model field is optional in both single and batch analysis requests,
defaulting to the server-configured MODEL env var. The model used
is recorded in the analysis result and database.

- Add model parameter to LLMAnalyzer.analyze_patent_content and
  analyze_patent_portfolio
- Add model field to CompanyAnalysisResult and API response
- Add model field to BatchAnalysisRequest
- Add GET /models endpoint listing supported models and the default
- Store model in llm_messages metadata for attribution

Closes leeworks-agents/SPARC#37

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

SPARC/api.py

@@ -41,6 +41,7 @@ class CompanyAnalysisResponse(BaseModel):
     patent_count: int
     success: bool
     error: str | None = None
+    model: str | None = None
     timestamp: datetime
 
 
@@ -54,6 +55,15 @@ class BatchAnalysisResponse(BaseModel):
     timestamp: datetime
 
 
+class CompanyAnalysisRequest(BaseModel):
+    """Request model for single company analysis with optional model selection."""
+
+    model: str | None = Field(
+        default=None,
+        description="LLM model to use (e.g. 'anthropic/claude-3.5-sonnet', 'openai/gpt-4o'). Defaults to server config.",
+    )
+
+
 class BatchAnalysisRequest(BaseModel):
     """Request model for batch company analysis."""
 
@@ -63,6 +73,10 @@ class BatchAnalysisRequest(BaseModel):
     max_workers: int = Field(
         default=3, ge=1, le=5, description="Max concurrent analyses"
     )
+    model: str | None = Field(
+        default=None,
+        description="LLM model to use for all analyses in this batch. Defaults to server config.",
+    )
 
 
 class JobStatus(BaseModel):
@@ -133,6 +147,7 @@ def _convert_result(result: CompanyAnalysisResult) -> CompanyAnalysisResponse:
         patent_count=result.patent_count,
         success=result.success,
         error=result.error,
+        model=result.model,
         timestamp=result.timestamp,
     )
 
@@ -389,6 +404,32 @@ async def get_analytics(
     )
 
 
+# ============== Model Selection Endpoints ==============
+
+# Supported models via OpenRouter
+SUPPORTED_MODELS = [
+    {"id": "anthropic/claude-3.5-sonnet", "name": "Claude 3.5 Sonnet", "provider": "Anthropic"},
+    {"id": "openai/gpt-4o", "name": "GPT-4o", "provider": "OpenAI"},
+    {"id": "openai/gpt-4o-mini", "name": "GPT-4o Mini", "provider": "OpenAI"},
+    {"id": "google/gemini-pro-1.5", "name": "Gemini Pro 1.5", "provider": "Google"},
+    {"id": "meta-llama/llama-3.1-70b-instruct", "name": "Llama 3.1 70B", "provider": "Meta"},
+]
+
+
+@app.get("/models", tags=["System"])
+async def list_models():
+    """List supported LLM models for analysis.
+
+    Returns the available models that can be passed as the `model` field
+    in analysis requests. The default model is determined by the `MODEL`
+    environment variable on the server.
+    """
+    return {
+        "models": SUPPORTED_MODELS,
+        "default": config.model,
+    }
+
+
 # ============== System Endpoints ==============
 
 

SPARC/llm.py

@@ -40,12 +40,13 @@ class LLMAnalyzer:
         else:
             self.client = None
 
-    def analyze_patent_content(self, patent_content: str, company_name: str) -> str:
+    def analyze_patent_content(self, patent_content: str, company_name: str, model: str | None = None) -> str:
         """Analyze patent content to estimate company innovation and performance.
 
         Args:
           patent_content: Minimized patent text (abstract, claims, summary)
           company_name: Name of the company for context
+          model: Optional model override (e.g. "openai/gpt-4o"). Defaults to config.
 
         Returns:
           Analysis text describing innovation quality and potential impact
@@ -63,6 +64,8 @@ Patent Content:
 
 Provide a concise analysis (2-3 paragraphs) focusing on what this patent reveals about the company's technical direction and competitive advantage."""
 
+        effective_model = model or self.model
+
         if self.test_mode:
             logger.debug("TEST MODE - Prompt that would be sent to LLM:\n%s", prompt)
             return "[TEST MODE - No API call made]"
@@ -81,7 +84,7 @@ Provide a concise analysis (2-3 paragraphs) focusing on what this patent reveals
                     response=cached["response"],
                     company_name=company_name,
                     analysis_type="single_patent",
-                    model=self.model,
+                    model=effective_model,
                     metadata={
                         "patent_content_length": len(patent_content),
                         "cache_hit": True,
@@ -94,7 +97,7 @@ Provide a concise analysis (2-3 paragraphs) focusing on what this patent reveals
         # Call API if no cache hit and client is available
         if self.client:
             response = self.client.chat.completions.create(
-                model=self.model,
+                model=effective_model,
                 max_tokens=1024,
                 messages=[{"role": "user", "content": prompt}],
             )
@@ -106,7 +109,7 @@ Provide a concise analysis (2-3 paragraphs) focusing on what this patent reveals
                 response=response_text,
                 company_name=company_name,
                 analysis_type="single_patent",
-                model=self.model,
+                model=effective_model,
                 metadata={"patent_content_length": len(patent_content)},
                 token_usage={
                     "prompt_tokens": response.usage.prompt_tokens,
@@ -124,13 +127,13 @@ Provide a concise analysis (2-3 paragraphs) focusing on what this patent reveals
             response=placeholder,
             company_name=company_name,
             analysis_type="single_patent",
-            model=self.model,
+            model=effective_model,
             metadata={"patent_content_length": len(patent_content), "pending": True}
         )
         return placeholder
 
     def analyze_patent_portfolio(
-        self, patents_data: list[Dict[str, str]], company_name: str
+        self, patents_data: list[Dict[str, str]], company_name: str, model: str | None = None
     ) -> str:
         """Analyze multiple patents to estimate overall company performance.
 
@@ -165,13 +168,16 @@ Patent Portfolio:
 
 Provide a comprehensive analysis (4-5 paragraphs) with a final verdict on the company's innovation strength and performance outlook."""
 
+        effective_model = model or self.model
+
         if self.test_mode:
             logger.debug("TEST MODE - Portfolio prompt:\n%s", prompt)
             return "[TEST MODE]"
 
         metadata = {
             "patent_count": len(patents_data),
-            "patent_ids": [p['patent_id'] for p in patents_data]
+            "patent_ids": [p['patent_id'] for p in patents_data],
+            "model": effective_model,
         }
 
         # Check cache first
@@ -188,7 +194,7 @@ Provide a comprehensive analysis (4-5 paragraphs) with a final verdict on the co
                     response=cached["response"],
                     company_name=company_name,
                     analysis_type="portfolio",
-                    model=self.model,
+                    model=effective_model,
                     metadata={
                         **metadata,
                         "cache_hit": True,
@@ -202,7 +208,7 @@ Provide a comprehensive analysis (4-5 paragraphs) with a final verdict on the co
         if self.client:
             try:
                 response = self.client.chat.completions.create(
-                    model=self.model,
+                    model=effective_model,
                     max_tokens=2048,
                     messages=[{"role": "user", "content": prompt}],
                 )
@@ -215,7 +221,7 @@ Provide a comprehensive analysis (4-5 paragraphs) with a final verdict on the co
                     response=response_text,
                     company_name=company_name,
                     analysis_type="portfolio",
-                    model=self.model,
+                    model=effective_model,
                     metadata=metadata,
                     token_usage={
                         "prompt_tokens": response.usage.prompt_tokens,
@@ -235,7 +241,7 @@ Provide a comprehensive analysis (4-5 paragraphs) with a final verdict on the co
             response=placeholder,
             company_name=company_name,
             analysis_type="portfolio",
-            model=self.model,
+            model=effective_model,
             metadata={**metadata, "pending": True}
         )
         return placeholder

SPARC/types.py

@@ -24,6 +24,7 @@ class CompanyAnalysisResult:
     patent_count: int
     success: bool
     error: str | None = None
+    model: str | None = None
     timestamp: datetime = field(default_factory=datetime.now)