remdb 0.3.0__py3-none-any.whl → 0.3.127__py3-none-any.whl

rem/services/rem/parser.py

@@ -50,9 +50,36 @@ class RemQueryParser:
         params: Dict[str, Any] = {}
         positional_args: List[str] = []
 
-        # Process remaining tokens
-        for token in tokens[1:]:
-            if "=" in token:
+        # For SQL queries, preserve the raw query (keywords like LIMIT are SQL keywords)
+        if query_type == QueryType.SQL:
+            # Everything after "SQL" is the raw SQL query
+            raw_sql = query_string[3:].strip()  # Skip "SQL" prefix
+            params["raw_query"] = raw_sql
+            return query_type, params
+
+        # Process remaining tokens, handling REM keywords
+        i = 1
+        while i < len(tokens):
+            token = tokens[i]
+            token_upper = token.upper()
+
+            # Handle REM keywords that take a value
+            if token_upper in ("LIMIT", "DEPTH", "THRESHOLD", "TYPE", "FROM", "WITH"):
+                if i + 1 < len(tokens):
+                    keyword_map = {
+                        "LIMIT": "limit",
+                        "DEPTH": "max_depth",
+                        "THRESHOLD": "threshold",
+                        "TYPE": "edge_types",
+                        "FROM": "initial_query",
+                        "WITH": "initial_query",
+                    }
+                    key = keyword_map[token_upper]
+                    value = tokens[i + 1]
+                    params[key] = self._convert_value(key, value)
+                    i += 2
+                    continue
+            elif "=" in token:
                 # It's a keyword argument
                 key, value = token.split("=", 1)
                 # Handle parameter aliases
@@ -61,6 +88,7 @@ class RemQueryParser:
             else:
                 # It's a positional argument part
                 positional_args.append(token)
+            i += 1
 
         # Map positional arguments to specific fields based on QueryType
         self._map_positional_args(query_type, positional_args, params)
@@ -133,13 +161,20 @@ class RemQueryParser:
             params["query_text"] = combined_value
 
         elif query_type == QueryType.SEARCH:
-            params["query_text"] = combined_value
+            # SEARCH expects: SEARCH <table> <query_text> [LIMIT n]
+            # First positional arg is table name, rest is query text
+            if len(positional_args) >= 2:
+                params["table_name"] = positional_args[0]
+                params["query_text"] = " ".join(positional_args[1:])
+            elif len(positional_args) == 1:
+                # Could be table name or query text - assume query text if no table
+                params["query_text"] = positional_args[0]
+            # If no positional args, params stays empty
 
         elif query_type == QueryType.TRAVERSE:
             params["initial_query"] = combined_value
         
-        # SQL typically requires named arguments (table=...), but if we supported 
-        # SQL SELECT * FROM ..., we might handle it differently. 
-        # For now, RemService expects table=...
-        # If there are positional args for SQL, we might ignore or raise, 
-        # but current service doesn't use them.
+        elif query_type == QueryType.SQL:
+            # SQL with positional args means "SQL SELECT * FROM ..." form
+            # Treat the combined positional args as the raw SQL query
+            params["raw_query"] = combined_value

rem/services/rem/service.py

@@ -13,6 +13,31 @@ Design:
 - All queries pushed down to Postgres for performance
 - Model schema inspection for validation only
 - Exceptions for missing fields/embeddings
+
+TODO: Staged Plan Execution
+- Implement execute_staged_plan() method for multi-stage query execution
+- Each stage can be:
+  1. Static query (query field): Execute REM dialect directly
+  2. Dynamic query (intent field): LLM interprets intent + previous results to build query
+- Flow for dynamic stages:
+  1. Gather results from depends_on stages (from previous_results or current execution)
+  2. Pass intent + previous results to LLM (like ask_rem but with context)
+  3. LLM generates REM query based on what it learned from previous stages
+  4. Execute generated query
+  5. Store results in stage_results for client to use in continuation
+- Multi-turn continuation:
+  - Client passes previous_results back from response's stage_results
+  - Client sets resume_from_stage to skip already-executed stages
+  - Server uses previous_results as context for depends_on lookups
+- Use cases:
+  - LOOKUP "Sarah" → intent: "find her team members" (LLM sees Sarah's graph_edges, builds TRAVERSE)
+  - SEARCH "API docs" → intent: "get authors" (LLM extracts author refs, builds LOOKUP)
+  - Complex graph exploration with LLM-driven navigation
+- API: POST /api/v1/query with:
+  - mode="staged-plan"
+  - plan=[{stage, query|intent, name, depends_on}]
+  - previous_results=[{stage, name, query_executed, results, count}] (for continuation)
+  - resume_from_stage=N (to skip completed stages)
 """
 
 from typing import Any
@@ -309,17 +334,26 @@ class RemService:
         )
 
         # Execute vector search via rem_search() PostgreSQL function
+        min_sim = params.min_similarity if params.min_similarity is not None else 0.3
+        limit = params.limit or 10
         query_params = get_search_params(
             query_embedding,
             table_name,
             field_name,
             tenant_id,
             provider,
-            params.min_similarity or 0.7,
-            params.limit or 10,
+            min_sim,
+            limit,
             tenant_id, # Use tenant_id (query.user_id) as user_id
         )
+        logger.debug(
+            f"SEARCH params: table={table_name}, field={field_name}, "
+            f"tenant_id={tenant_id}, provider={provider}, "
+            f"min_similarity={min_sim}, limit={limit}, "
+            f"embedding_dims={len(query_embedding)}"
+        )
         results = await self.db.execute(SEARCH_QUERY, query_params)
+        logger.debug(f"SEARCH results: {len(results)} rows")
 
         return {
             "query_type": "SEARCH",

rem/services/session/compression.py

@@ -14,6 +14,21 @@ from typing import Any
 
 from loguru import logger
 
+# Max length for entity keys (kv_store.entity_key is varchar(255))
+MAX_ENTITY_KEY_LENGTH = 255
+
+
+def truncate_key(key: str, max_length: int = MAX_ENTITY_KEY_LENGTH) -> str:
+    """Truncate a key to max length, preserving useful suffix if possible."""
+    if len(key) <= max_length:
+        return key
+    # Keep first part and add hash suffix for uniqueness
+    import hashlib
+    hash_suffix = hashlib.md5(key.encode()).hexdigest()[:8]
+    truncated = key[:max_length - 9] + "-" + hash_suffix
+    logger.warning(f"Truncated key from {len(key)} to {len(truncated)} chars: {key[:50]}...")
+    return truncated
+
 from rem.models.entities import Message
 from rem.services.postgres import PostgresService, Repository
 from rem.settings import settings
@@ -151,15 +166,20 @@ class SessionMessageStore:
             return f"msg-{message_index}"
 
         # Create entity key for REM LOOKUP: session-{session_id}-msg-{index}
-        entity_key = f"session-{session_id}-msg-{message_index}"
+        # Truncate to avoid exceeding kv_store.entity_key varchar(255) limit
+        entity_key = truncate_key(f"session-{session_id}-msg-{message_index}")
 
         # Create Message entity for assistant response
+        # Use pre-generated id from message dict if available (for frontend feedback)
         msg = Message(
+            id=message.get("id"),  # Use pre-generated ID if provided
             content=message.get("content", ""),
             message_type=message.get("role", "assistant"),
             session_id=session_id,
             tenant_id=self.user_id,  # Set tenant_id to user_id (application scoped to user)
             user_id=user_id or self.user_id,
+            trace_id=message.get("trace_id"),
+            span_id=message.get("span_id"),
             metadata={
                 "message_index": message_index,
                 "entity_key": entity_key,  # Store entity key for LOOKUP
@@ -268,11 +288,14 @@ class SessionMessageStore:
                 # Short assistant messages, user messages, and system messages stored as-is
                 # Store ALL messages in database for full audit trail
                 msg = Message(
+                    id=message.get("id"),  # Use pre-generated ID if provided
                     content=content,
                     message_type=message.get("role", "user"),
                     session_id=session_id,
                     tenant_id=self.user_id,  # Set tenant_id to user_id (application scoped to user)
                     user_id=user_id or self.user_id,
+                    trace_id=message.get("trace_id"),
+                    span_id=message.get("span_id"),
                     metadata={
                         "message_index": idx,
                         "timestamp": message.get("timestamp"),

rem/services/session/reload.py

@@ -65,7 +65,7 @@ async def reload_session(
             session_id=session_id, user_id=user_id, decompress=decompress_messages
         )
 
-        logger.info(
+        logger.debug(
             f"Reloaded {len(messages)} messages for session {session_id} "
             f"(decompressed={decompress_messages})"
         )

rem/services/user_service.py

@@ -0,0 +1,98 @@
+"""
+User Service - User account management.
+
+Handles user creation, profile updates, and session linking.
+"""
+
+from datetime import datetime
+from typing import Optional
+
+from loguru import logger
+
+from ..models.entities.user import User, UserTier
+from .postgres.repository import Repository
+from .postgres.service import PostgresService
+
+
+class UserService:
+    """
+    Service for managing user accounts and sessions.
+    """
+
+    def __init__(self, db: PostgresService):
+        self.db = db
+        self.repo = Repository(User, "users", db=db)
+
+    async def get_or_create_user(
+        self,
+        email: str,
+        tenant_id: str = "default",
+        name: str = "New User",
+        avatar_url: Optional[str] = None,
+    ) -> User:
+        """
+        Get existing user by email or create a new one.
+        """
+        users = await self.repo.find(filters={"email": email}, limit=1)
+        
+        if users:
+            user = users[0]
+            # Update profile if needed (e.g., name/avatar from OAuth)
+            updated = False
+            if name and user.name == "New User": # Only update if placeholder
+                user.name = name
+                updated = True
+            
+            # Store avatar in metadata if provided
+            if avatar_url:
+                user.metadata = user.metadata or {}
+                if user.metadata.get("avatar_url") != avatar_url:
+                    user.metadata["avatar_url"] = avatar_url
+                    updated = True
+            
+            if updated:
+                user.updated_at = datetime.utcnow()
+                await self.repo.upsert(user)
+            
+            return user
+        
+        # Create new user
+        user = User(
+            tenant_id=tenant_id,
+            user_id=email, # Use email as user_id for now? Or UUID?
+            # The User model has 'user_id' field but also 'id' UUID.
+            # Usually user_id is the external ID or email.
+            name=name,
+            email=email,
+            tier=UserTier.FREE,
+            created_at=datetime.utcnow(),
+            updated_at=datetime.utcnow(),
+            metadata={"avatar_url": avatar_url} if avatar_url else {},
+        )
+        await self.repo.upsert(user)
+        logger.info(f"Created new user: {email}")
+        return user
+
+    async def link_anonymous_session(self, user: User, anon_id: str) -> None:
+        """
+        Link an anonymous session ID to a user account.
+        
+        This allows merging history from the anonymous session into the user's profile.
+        """
+        if not anon_id:
+            return
+
+        # Check if already linked
+        if anon_id in user.anonymous_ids:
+            return
+
+        # Add to list
+        user.anonymous_ids.append(anon_id)
+        user.updated_at = datetime.utcnow()
+        
+        # Save
+        await self.repo.upsert(user)
+        logger.info(f"Linked anonymous session {anon_id} to user {user.email}")
+        
+        # TODO: Migrate/Merge actual data (rate limit counts, history) if needed.
+        # For now, we just link the IDs so future queries can include data from this anon_id.