fix: Replace non-existent model names and handle Neo4j index race condition

- Replace hardcoded non-existent models (gpt-5-mini, gpt-5-nano, gpt-4.1)
  with valid gpt-4o-mini across config files and default client
- Add exception handling for EquivalentSchemaRuleAlreadyExists in
  Neo4j index creation to prevent race condition failures on startup

graphiti_core/driver/neo4j_driver.py

@@ -19,6 +19,7 @@ from collections.abc import Coroutine
 from typing import Any
 
 from neo4j import AsyncGraphDatabase, EagerResult
+from neo4j.exceptions import ClientError
 from typing_extensions import LiteralString
 
 from graphiti_core.driver.driver import GraphDriver, GraphDriverSession, GraphProvider
@@ -88,6 +89,21 @@ class Neo4jDriver(GraphDriver):
             'CALL db.indexes() YIELD name DROP INDEX name',
         )
 
+    async def _execute_index_query(self, query: LiteralString) -> EagerResult | None:
+        """Execute an index creation query, ignoring 'index already exists' errors.
+
+        Neo4j can raise EquivalentSchemaRuleAlreadyExists when concurrent CREATE INDEX
+        IF NOT EXISTS queries race, even though the index exists. This is safe to ignore.
+        """
+        try:
+            return await self.execute_query(query)
+        except ClientError as e:
+            # Ignore "equivalent index already exists" error (race condition with IF NOT EXISTS)
+            if 'EquivalentSchemaRuleAlreadyExists' in str(e):
+                logger.debug(f'Index already exists (concurrent creation): {query[:50]}...')
+                return None
+            raise
+
     async def build_indices_and_constraints(self, delete_existing: bool = False):
         if delete_existing:
             await self.delete_all_indexes()
@@ -98,14 +114,7 @@ class Neo4jDriver(GraphDriver):
 
         index_queries: list[LiteralString] = range_indices + fulltext_indices
 
-        await semaphore_gather(
-            *[
-                self.execute_query(
-                    query,
-                )
-                for query in index_queries
-            ]
-        )
+        await semaphore_gather(*[self._execute_index_query(query) for query in index_queries])
 
     async def health_check(self) -> None:
         """Check Neo4j connectivity by running the driver's verify_connectivity method."""

graphiti_core/llm_client/openai_base_client.py

@@ -31,8 +31,8 @@ from .errors import RateLimitError, RefusalError
 
 logger = logging.getLogger(__name__)
 
-DEFAULT_MODEL = 'gpt-5-mini'
-DEFAULT_SMALL_MODEL = 'gpt-5-nano'
+DEFAULT_MODEL = 'gpt-4o-mini'
+DEFAULT_SMALL_MODEL = 'gpt-4o-mini'
 DEFAULT_REASONING = 'minimal'
 DEFAULT_VERBOSITY = 'low'
 

mcp_server/config/config-docker-falkordb-combined.yaml

@@ -8,7 +8,7 @@ server:
 
 llm:
   provider: "openai"  # Options: openai, azure_openai, anthropic, gemini, groq
-  model: "gpt-5-mini"
+  model: "gpt-4o-mini"
   max_tokens: 4096
 
   providers:

mcp_server/config/config-docker-falkordb.yaml

@@ -8,7 +8,7 @@ server:
   
 llm:
   provider: "openai"  # Options: openai, azure_openai, anthropic, gemini, groq
-  model: "gpt-5-mini"
+  model: "gpt-4o-mini"
   max_tokens: 4096
   
   providers:

mcp_server/config/config-docker-neo4j.yaml

@@ -8,7 +8,7 @@ server:
   
 llm:
   provider: "openai"  # Options: openai, azure_openai, anthropic, gemini, groq
-  model: "gpt-5-mini"
+  model: "gpt-4o-mini"
   max_tokens: 4096
   
   providers:

mcp_server/config/config.yaml

@@ -12,7 +12,7 @@ server:
   
 llm:
   provider: "openai"  # Options: openai, azure_openai, anthropic, gemini, groq
-  model: "gpt-5-mini"
+  model: "gpt-4o-mini"
   max_tokens: 4096
   
   providers:

mcp_server/src/config/schema.py

@@ -147,7 +147,7 @@ class LLMConfig(BaseModel):
     """LLM configuration."""
 
     provider: str = Field(default='openai', description='LLM provider')
-    model: str = Field(default='gpt-4.1', description='Model name')
+    model: str = Field(default='gpt-4o-mini', description='Model name')
     temperature: float | None = Field(
         default=None, description='Temperature (optional, defaults to None for reasoning models)'
     )