fix: address CodeRabbit review comments for LLM registry PR

- Add try/except around startup LLM registry refresh (rest_api.py)
- Fix race condition in refresh_llm_costs - build list then swap (block_cost_config.py)
- Add @lru_cache to _get_llm_models() with cache clear on refresh (db.py, llm_routes.py)
- Fix retry using wrong model's max_output_tokens after fallback (llm.py)
- Remove redundant @@index([slug]) from schema.prisma
- Fix operationId collision in openapi.json

autogpt_platform/backend/backend/api/features/admin/llm_routes.py

@@ -47,6 +47,8 @@ async def _refresh_runtime_state() -> None:
             logger.info("Cleared v2 builder providers cache")
             builder_db._build_cached_search_results.cache_clear()
             logger.info("Cleared v2 builder search results cache")
+            builder_db._get_llm_models.cache_clear()
+            logger.info("Cleared v2 builder LLM models cache")
         except Exception as e:
             logger.debug("Could not clear v2 builder cache: %s", e)
 

autogpt_platform/backend/backend/api/features/builder/db.py

@@ -1,6 +1,7 @@
 import logging
 from dataclasses import dataclass
 from difflib import SequenceMatcher
+from functools import lru_cache
 from typing import Any, Sequence, get_args, get_origin
 
 import prisma
@@ -40,11 +41,17 @@ from .model import (
 logger = logging.getLogger(__name__)
 
 
-def _get_llm_models() -> list[str]:
-    """Get LLM model names for search matching from the registry."""
-    return [
+@lru_cache(maxsize=1)
+def _get_llm_models() -> tuple[str, ...]:
+    """Get LLM model names for search matching from the registry.
+
+    Cached to avoid rebuilding on every search call.
+    Cache is cleared when registry is refreshed via _refresh_runtime_state.
+    Returns tuple for hashability (required by lru_cache).
+    """
+    return tuple(
         slug.lower().replace("-", " ") for slug in get_all_model_slugs_for_validation()
-    ]
+    )
 
 
 MAX_LIBRARY_AGENT_RESULTS = 100

autogpt_platform/backend/backend/api/rest_api.py

@@ -120,8 +120,11 @@ async def lifespan_context(app: fastapi.FastAPI):
     AutoRegistry.patch_integrations()
 
     # Refresh LLM registry before initializing blocks so blocks can use registry data
-    await llm_registry.refresh_llm_registry()
-    await refresh_llm_costs()
+    try:
+        await llm_registry.refresh_llm_registry()
+        await refresh_llm_costs()
+    except Exception as e:
+        logger.warning(f"Failed to refresh LLM registry/costs at startup: {e}")
 
     # Clear block schema caches so they're regenerated with updated discriminator_mapping
     from backend.blocks._base import BlockSchema

autogpt_platform/backend/backend/blocks/llm.py

@@ -378,6 +378,9 @@ class LLMResponse(BaseModel):
     prompt_tokens: int
     completion_tokens: int
     reasoning: Optional[str] = None
+    resolved_max_output_tokens: Optional[int] = (
+        None  # Max output tokens of resolved model (after fallback)
+    )
 
 
 def convert_openai_tool_fmt_to_anthropic(
@@ -553,6 +556,7 @@ async def llm_call(
             prompt_tokens=response.usage.prompt_tokens if response.usage else 0,
             completion_tokens=response.usage.completion_tokens if response.usage else 0,
             reasoning=reasoning,
+            resolved_max_output_tokens=model_max_output,
         )
     elif provider == "anthropic":
 
@@ -633,6 +637,7 @@ async def llm_call(
                 prompt_tokens=resp.usage.input_tokens,
                 completion_tokens=resp.usage.output_tokens,
                 reasoning=reasoning,
+                resolved_max_output_tokens=model_max_output,
             )
         except anthropic.APIError as e:
             error_message = f"Anthropic API error: {str(e)}"
@@ -658,6 +663,7 @@ async def llm_call(
             prompt_tokens=response.usage.prompt_tokens if response.usage else 0,
             completion_tokens=response.usage.completion_tokens if response.usage else 0,
             reasoning=None,
+            resolved_max_output_tokens=model_max_output,
         )
     elif provider == "ollama":
         if tools:
@@ -680,6 +686,7 @@ async def llm_call(
             prompt_tokens=response.get("prompt_eval_count") or 0,
             completion_tokens=response.get("eval_count") or 0,
             reasoning=None,
+            resolved_max_output_tokens=model_max_output,
         )
     elif provider == "open_router":
         tools_param = tools if tools else openai.NOT_GIVEN
@@ -722,6 +729,7 @@ async def llm_call(
             prompt_tokens=response.usage.prompt_tokens if response.usage else 0,
             completion_tokens=response.usage.completion_tokens if response.usage else 0,
             reasoning=reasoning,
+            resolved_max_output_tokens=model_max_output,
         )
     elif provider == "llama_api":
         tools_param = tools if tools else openai.NOT_GIVEN
@@ -764,6 +772,7 @@ async def llm_call(
             prompt_tokens=response.usage.prompt_tokens if response.usage else 0,
             completion_tokens=response.usage.completion_tokens if response.usage else 0,
             reasoning=reasoning,
+            resolved_max_output_tokens=model_max_output,
         )
     elif provider == "aiml_api":
         client = openai.AsyncOpenAI(
@@ -792,6 +801,7 @@ async def llm_call(
                 completion.usage.completion_tokens if completion.usage else 0
             ),
             reasoning=None,
+            resolved_max_output_tokens=model_max_output,
         )
     elif provider == "v0":
         tools_param = tools if tools else openai.NOT_GIVEN
@@ -828,6 +838,7 @@ async def llm_call(
             prompt_tokens=response.usage.prompt_tokens if response.usage else 0,
             completion_tokens=response.usage.completion_tokens if response.usage else 0,
             reasoning=reasoning,
+            resolved_max_output_tokens=model_max_output,
         )
     else:
         raise ValueError(f"Unsupported LLM provider: {provider}")
@@ -953,6 +964,7 @@ class AIStructuredResponseGeneratorBlock(AIBlockBase):
                     prompt_tokens=0,
                     completion_tokens=0,
                     reasoning=None,
+                    resolved_max_output_tokens=None,
                 ),
                 "get_collision_proof_output_tag_id": lambda *args: "test123456",
             },
@@ -1028,6 +1040,11 @@ class AIStructuredResponseGeneratorBlock(AIBlockBase):
         error_feedback_message = ""
         llm_model = input_data.model
 
+        # Resolve the model once before retry loop to get correct max_output_tokens
+        # for retry logic (handles fallback models correctly)
+        resolved = await resolve_model_for_call(llm_model)
+        resolved_max_output = resolved.max_output_tokens
+
         for retry_count in range(input_data.retry):
             logger.debug(f"LLM request: {prompt}")
             try:
@@ -1150,7 +1167,8 @@ class AIStructuredResponseGeneratorBlock(AIBlockBase):
                     or "token limit" in str(e).lower()
                 ):
                     if input_data.max_tokens is None:
-                        input_data.max_tokens = llm_model.max_output_tokens or 4096
+                        # Use resolved model's max_output_tokens (handles fallback correctly)
+                        input_data.max_tokens = resolved_max_output or 4096
                     input_data.max_tokens = int(input_data.max_tokens * 0.85)
                     logger.debug(
                         f"Reducing max_tokens to {input_data.max_tokens} for next attempt"

autogpt_platform/backend/backend/data/block_cost_config.py

@@ -163,8 +163,11 @@ async def refresh_llm_costs() -> None:
     This function also checks for active model migrations with custom pricing overrides
     and applies them to ensure accurate billing.
     """
+    # Build new costs first, then swap atomically to avoid race condition
+    # where concurrent readers see an empty list during the await
+    new_costs = await _build_llm_costs_from_registry()
     LLM_COST.clear()
-    LLM_COST.extend(await _build_llm_costs_from_registry())
+    LLM_COST.extend(new_costs)
 
 
 # Initial load will happen after registry is refreshed at startup

autogpt_platform/backend/schema.prisma

@@ -1265,7 +1265,7 @@ model LlmModel {
 
   @@index([providerId, isEnabled])
   @@index([creatorId])
-  @@index([slug])
+  // Note: slug already has @unique which creates an implicit index
 }
 
 model LlmModelCost {
@@ -1330,6 +1330,9 @@ model LlmModelMigration {
   @@index([isReverted])
   @@index([sourceModelSlug, isReverted]) // Composite index for active migration queries
 }
+
+////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////
 //////////////   OAUTH PROVIDER TABLES    //////////////////
 ////////////////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////

autogpt_platform/frontend/src/app/api/openapi.json

@@ -4636,7 +4636,7 @@
         "tags": ["v2", "admin", "llm", "llm", "admin"],
         "summary": "Get creator details",
         "description": "Get details of a specific model creator.",
-        "operationId": "getV2Get creator details",
+        "operationId": "getV2GetLlmCreatorDetails",
         "security": [{ "HTTPBearerJWT": [] }],
         "parameters": [
           {