fix iuo

benchmarks/common/benchmark_runner.py

@@ -718,25 +718,23 @@ class BenchmarkRunner:
             await self.memory.delete_agent(agent_id)
             console.print(f"    [green]✓[/green] Cleared agent data")
 
-            with Progress(
+            # Collect all sessions from all items into one large batch
-                SpinnerColumn(),
+            console.print(f"    [yellow]Collecting sessions from all items...[/yellow]")
-                TextColumn("[progress.description]{task.description}"),
+            all_sessions = []
-                BarColumn(),
+            for item in items:
-                TextColumn("[progress.percentage]{task.percentage:>3.0f}%"),
+                item_sessions = self.dataset.prepare_sessions_for_ingestion(item)
-                console=console
+                all_sessions.extend(item_sessions)
-            ) as progress:
+
-                task = progress.add_task(
+            console.print(f"    [cyan]Collected {len(all_sessions)} sessions from {len(items)} items[/cyan]")
-                    f"[cyan]Ingesting {len(items)} items...",
+            console.print(f"    [yellow]Ingesting in one batch (auto-chunks if needed)...[/yellow]")
-                    total=len(items)
+
+            # Ingest all sessions in one batch call (will auto-chunk if too large)
+            await self.memory.put_batch_async(
+                agent_id=agent_id,
+                contents=all_sessions
             )
 
-                total_sessions = 0
+            console.print(f"    [green]✓[/green] Ingested {len(all_sessions)} sessions from {len(items)} items")
-                for item in items:
-                    num_sessions = await self.ingest_conversation(item, agent_id)
-                    total_sessions += num_sessions
-                    progress.update(task, advance=1)
-
-            console.print(f"    [green]✓[/green] Ingested {total_sessions} sessions from {len(items)} items")
         else:
             console.print(f"\n[3] Skipping ingestion (using existing data)")
 

memora/fact_extraction.py

@@ -423,8 +423,6 @@ Remember:
             # Convert to dict format
             chunk_facts = [fact.model_dump() for fact in extraction_response.facts]
 
-            logger.info(f"          [1.3.{chunk_index + 1}] Chunk {chunk_index + 1}/{total_chunks} LLM call: {len(chunk_facts)} facts from {len(chunk)} chars in {llm_call_time:.3f}s")
-
             return chunk_facts
 
         except BadRequestError as e:

memora/temporal_semantic_memory.py

@@ -476,6 +476,7 @@ class TemporalSemanticMemory(
         - Extracts facts from all contents in parallel
         - Generates ALL embeddings in ONE batch
         - Does ALL database operations in ONE transaction
+        - Automatically chunks large batches to prevent timeouts
 
         Args:
             agent_id: Unique identifier for the agent
@@ -505,15 +506,102 @@ class TemporalSemanticMemory(
             # Returns: [["unit-id-1"], ["unit-id-2"]]
         """
         start_time = time.time()
-        log_buffer = []  # Buffer all logs to avoid interleaving
-        log_buffer.append(f"{'='*60}")
-        log_buffer.append(f"PUT_BATCH_ASYNC START: {agent_id}")
-        log_buffer.append(f"Batch size: {len(contents)} content items")
-        log_buffer.append(f"{'='*60}")
 
         if not contents:
             return []
 
+        # Auto-chunk large batches by character count to avoid timeouts and memory issues
+        # Calculate total character count
+        total_chars = sum(len(item.get("content", "")) for item in contents)
+
+        # Threshold: 50,000 characters per sub-batch (roughly 12k tokens with average 4 chars/token)
+        CHARS_PER_BATCH = 50_000
+
+        if total_chars > CHARS_PER_BATCH:
+            # Split into smaller batches based on character count
+            logger.info(f"Large batch detected ({total_chars:,} chars from {len(contents)} items). Splitting into sub-batches of ~{CHARS_PER_BATCH:,} chars each...")
+
+            sub_batches = []
+            current_batch = []
+            current_batch_chars = 0
+
+            for item in contents:
+                item_chars = len(item.get("content", ""))
+
+                # If adding this item would exceed the limit, start a new batch
+                # (unless current batch is empty - then we must include it even if it's large)
+                if current_batch and current_batch_chars + item_chars > CHARS_PER_BATCH:
+                    sub_batches.append(current_batch)
+                    current_batch = [item]
+                    current_batch_chars = item_chars
+                else:
+                    current_batch.append(item)
+                    current_batch_chars += item_chars
+
+            # Add the last batch
+            if current_batch:
+                sub_batches.append(current_batch)
+
+            logger.info(f"Split into {len(sub_batches)} sub-batches: {[len(b) for b in sub_batches]} items each")
+
+            # Process each sub-batch using internal method (skip chunking check)
+            all_results = []
+            for i, sub_batch in enumerate(sub_batches, 1):
+                sub_batch_chars = sum(len(item.get("content", "")) for item in sub_batch)
+                logger.info(f"Processing sub-batch {i}/{len(sub_batches)}: {len(sub_batch)} items, {sub_batch_chars:,} chars")
+
+                sub_results = await self._put_batch_async_internal(
+                    agent_id=agent_id,
+                    contents=sub_batch,
+                    document_id=document_id,
+                    document_metadata=document_metadata,
+                    upsert=upsert and i == 1,  # Only upsert on first batch
+                    fact_type_override=fact_type_override,
+                    confidence_score=confidence_score
+                )
+                all_results.extend(sub_results)
+
+            total_time = time.time() - start_time
+            logger.info(f"PUT_BATCH_ASYNC (chunked) COMPLETE: {len(all_results)} results from {len(contents)} contents in {total_time:.3f}s")
+            return all_results
+
+        # Small batch - use internal method directly
+        return await self._put_batch_async_internal(
+            agent_id=agent_id,
+            contents=contents,
+            document_id=document_id,
+            document_metadata=document_metadata,
+            upsert=upsert,
+            fact_type_override=fact_type_override,
+            confidence_score=confidence_score
+        )
+
+    async def _put_batch_async_internal(
+        self,
+        agent_id: str,
+        contents: List[Dict[str, Any]],
+        document_id: Optional[str] = None,
+        document_metadata: Optional[Dict[str, Any]] = None,
+        upsert: bool = False,
+        fact_type_override: Optional[str] = None,
+        confidence_score: Optional[float] = None,
+    ) -> List[List[str]]:
+        """
+        Internal method for batch processing without chunking logic.
+
+        Assumes contents are already appropriately sized (< 50k chars).
+        Called by put_batch_async after chunking large batches.
+        """
+        start_time = time.time()
+        total_chars = sum(len(item.get("content", "")) for item in contents)
+
+        # Buffer all logs to avoid interleaving
+        log_buffer = []
+        log_buffer.append(f"{'='*60}")
+        log_buffer.append(f"PUT_BATCH_ASYNC START: {agent_id}")
+        log_buffer.append(f"Batch size: {len(contents)} content items, {total_chars:,} chars")
+        log_buffer.append(f"{'='*60}")
+
         # Step 1: Extract facts from ALL contents in parallel
         step_start = time.time()
 

memora/web/__init__.py

@@ -3,13 +3,6 @@ Web interface for memory system.
 
 Provides FastAPI app and visualization interface.
 """
-from .server import create_app
+from .server import app, create_app
-
-# Lazy import of app to avoid initialization issues
-def __getattr__(name):
-    if name == "app":
-        from .server import _get_app
-        return _get_app()
-    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
 __all__ = ["app", "create_app"]

memora/web/server.py

@@ -22,7 +22,7 @@ from memora.embeddings import Embeddings
 
 import logging
 
-logging.basicConfig(level=logging.INFO)
+
 
 # Environment variables are loaded by the shell script that calls this module
 # No need to load .env files here as they're sourced by start-server.sh
@@ -621,6 +621,7 @@ def _register_routes(app: FastAPI):
                 document_metadata=request.document_metadata,
                 upsert=request.upsert
             )
+            logging.info(f"Batch put result: {result}")
 
             return BatchPutResponse(
                 success=True,
@@ -662,33 +663,21 @@ def _register_routes(app: FastAPI):
 
 
 
-def _create_default_app():
+# Create app at module level (required for uvicorn import string)
-    """Create app instance with default environment configuration."""
+_memory = TemporalSemanticMemory(
-    _memory = TemporalSemanticMemory(
     db_url=os.getenv("DATABASE_URL"),
     memory_llm_provider=os.getenv("MEMORY_LLM_PROVIDER", "groq"),
     memory_llm_api_key=os.getenv("MEMORY_LLM_API_KEY"),
     memory_llm_model=os.getenv("MEMORY_LLM_MODEL", "openai/gpt-oss-120b"),
     memory_llm_base_url=os.getenv("MEMORY_LLM_BASE_URL") or None,
-    )
+)
-    return create_app(_memory)
+app = create_app(_memory)
-
-
-# Module-level app instance (lazy initialization)
-_app_instance = None
-
-
-def _get_app():
-    """Get or create the app instance."""
-    global _app_instance
-    if _app_instance is None:
-        _app_instance = _create_default_app()
-    return _app_instance
 
 
 if __name__ == "__main__":
     import uvicorn
     import argparse
+    logging.basicConfig(level=logging.INFO)
 
     # Parse CLI arguments
     parser = argparse.ArgumentParser(description="Memory Graph API Server")
@@ -708,9 +697,6 @@ if __name__ == "__main__":
 
     args = parser.parse_args()
 
-    # Create app after parsing args (so --help works without DB connection)
-    app = _create_default_app()
-
     print("\n" + "=" * 80)
     print("Memory Graph API Server")
     print("=" * 80)
@@ -721,9 +707,12 @@ if __name__ == "__main__":
     print(f"Log Level: {args.log_level}")
     print("=" * 80 + "\n")
 
+    # Always use import string for uvicorn (required for reload and workers)
+    app_ref = "memora.web.server:app"
+
     # Prepare uvicorn config
     uvicorn_config = {
-        "app": app,
+        "app": app_ref,
         "host": args.host,
         "port": args.port,
         "reload": args.reload,

memora/web/static/js/app.js

@@ -1127,7 +1127,7 @@ window.runSearchInPane = async function(paneId) {
         return;
     }
 
-    try:
+    try {
         // Prepare request body with optional fact_type
         const requestBody = {
             query: query,