remdb 0.2.6__py3-none-any.whl

rem/utils/model_helpers.py

@@ -0,0 +1,236 @@
+"""
+Pydantic Model Helper Utilities.
+
+Utilities for working with REM Pydantic models following our conventions:
+
+Business Key (entity_key) Detection:
+1. Field with json_schema_extra={"entity_key": True}
+2. Common business key fields: name, uri, key, label
+3. Fallback to "id" (unique by UUID only)
+
+Embedding Field Detection:
+1. Field with json_schema_extra={"embed": True}
+2. Common content fields: content, description, summary, etc.
+3. Explicit disable with json_schema_extra={"embed": False}
+
+Table Name Inference:
+1. model_config.json_schema_extra.table_name
+2. CamelCase → snake_case + pluralization
+"""
+
+from typing import Any, Type
+
+from loguru import logger
+from pydantic import BaseModel
+
+
+def get_entity_key_field(model: Type[BaseModel]) -> str:
+    """
+    Get the business key field for KV store lookups.
+
+    Follows REM conventions:
+    1. Field with json_schema_extra={"entity_key": True}
+    2. "name" field (most common for resources, moments, etc.)
+    3. "uri" field (for files)
+    4. "key" or "label" fields
+    5. Fallback to "id" (UUID only)
+
+    Args:
+        model: Pydantic model class
+
+    Returns:
+        Field name to use as entity_key
+
+    Example:
+        >>> from rem.models.entities import Resource
+        >>> get_entity_key_field(Resource)
+        'name'
+    """
+    # Check for explicit entity_key marker
+    for field_name, field_info in model.model_fields.items():
+        json_extra = getattr(field_info, "json_schema_extra", None)
+        if json_extra and isinstance(json_extra, dict):
+            if json_extra.get("entity_key") is True:
+                logger.debug(f"Using explicit entity_key field: {field_name}")
+                return field_name
+
+    # Check for common business key fields
+    for candidate in ["name", "uri", "key", "label", "title"]:
+        if candidate in model.model_fields:
+            logger.debug(f"Using conventional entity_key field: {candidate}")
+            return candidate
+
+    # Fallback to id (unique by UUID only)
+    logger.warning(
+        f"No business key found for {model.__name__}, using 'id' (UUID only)"
+    )
+    return "id"
+
+
+def get_table_name(model: Type[BaseModel]) -> str:
+    """
+    Get table name for a Pydantic model.
+
+    Follows REM conventions:
+    1. model_config.json_schema_extra.table_name (explicit)
+    2. CamelCase → snake_case + pluralization
+
+    Args:
+        model: Pydantic model class
+
+    Returns:
+        Table name
+
+    Example:
+        >>> from rem.models.entities import Resource
+        >>> get_table_name(Resource)
+        'resources'
+    """
+    import re
+
+    # Check for explicit table_name
+    if hasattr(model, "model_config"):
+        model_config = model.model_config
+        if isinstance(model_config, dict):
+            json_extra = model_config.get("json_schema_extra", {})
+            if isinstance(json_extra, dict) and "table_name" in json_extra:
+                return json_extra["table_name"]
+
+    # Infer from class name
+    name = model.__name__
+
+    # Convert CamelCase to snake_case
+    name = re.sub("(.)([A-Z][a-z]+)", r"\1_\2", name)
+    name = re.sub("([a-z0-9])([A-Z])", r"\1_\2", name).lower()
+
+    # Pluralize
+    if not name.endswith("s"):
+        if name.endswith("y"):
+            name = name[:-1] + "ies"  # category -> categories
+        else:
+            name = name + "s"  # resource -> resources
+
+    return name
+
+
+def get_embeddable_fields(model: Type[BaseModel]) -> list[str]:
+    """
+    Get list of fields that should have embeddings generated.
+
+    Follows REM conventions:
+    1. Field with json_schema_extra={"embed": True} → always embed
+    2. Field with json_schema_extra={"embed": False} → never embed
+    3. Common content fields → embed by default
+    4. Otherwise → don't embed
+
+    Args:
+        model: Pydantic model class
+
+    Returns:
+        List of field names to generate embeddings for
+
+    Example:
+        >>> from rem.models.entities import Resource
+        >>> fields = get_embeddable_fields(Resource)
+        >>> "content" in fields
+        True
+    """
+    # Common content fields that embed by default
+    DEFAULT_EMBED_FIELDS = {
+        "content",
+        "description",
+        "summary",
+        "text",
+        "body",
+        "message",
+        "notes",
+    }
+
+    embeddable = []
+
+    for field_name, field_info in model.model_fields.items():
+        # Check json_schema_extra for explicit embed configuration
+        json_extra = getattr(field_info, "json_schema_extra", None)
+        if json_extra and isinstance(json_extra, dict):
+            embed = json_extra.get("embed")
+            if embed is True:
+                embeddable.append(field_name)
+                continue
+            elif embed is False:
+                # Explicitly disabled
+                continue
+
+        # Check if field name matches common content fields
+        if field_name.lower() in DEFAULT_EMBED_FIELDS:
+            embeddable.append(field_name)
+
+    return embeddable
+
+
+def should_skip_field(field_name: str) -> bool:
+    """
+    Check if a field should be skipped during SQL generation.
+
+    System fields that are added separately:
+    - id (added as PRIMARY KEY)
+    - tenant_id (added for multi-tenancy)
+    - user_id (added for ownership)
+    - created_at, updated_at, deleted_at (added as system timestamps)
+    - graph_edges, metadata (added as JSONB system fields)
+    - tags, column (CoreModel fields)
+
+    Args:
+        field_name: Name of the field
+
+    Returns:
+        True if field should be skipped
+
+    Example:
+        >>> should_skip_field("id")
+        True
+        >>> should_skip_field("name")
+        False
+    """
+    SYSTEM_FIELDS = {
+        "id",
+        "tenant_id",
+        "user_id",
+        "created_at",
+        "updated_at",
+        "deleted_at",
+        "graph_edges",
+        "metadata",
+        "tags",
+        "column",
+    }
+
+    return field_name in SYSTEM_FIELDS
+
+
+def get_model_metadata(model: Type[BaseModel]) -> dict[str, Any]:
+    """
+    Extract REM-specific metadata from a Pydantic model.
+
+    Returns:
+        Dict with:
+        - table_name: Database table name
+        - entity_key_field: Business key field name
+        - embeddable_fields: List of fields to embed
+        - model_name: Original model class name
+
+    Example:
+        >>> from rem.models.entities import Resource
+        >>> meta = get_model_metadata(Resource)
+        >>> meta["table_name"]
+        'resources'
+        >>> meta["entity_key_field"]
+        'name'
+        >>> "content" in meta["embeddable_fields"]
+        True
+    """
+    return {
+        "model_name": model.__name__,
+        "table_name": get_table_name(model),
+        "entity_key_field": get_entity_key_field(model),
+        "embeddable_fields": get_embeddable_fields(model),
+    }

rem/utils/schema_loader.py

@@ -0,0 +1,229 @@
+"""
+Centralized schema loading utility for agent schemas.
+
+This module provides a single, consistent implementation for loading
+agent schemas from YAML files across the entire codebase (API, CLI, agent factory).
+
+Design Pattern:
+- Search standard locations: schemas/agents/, schemas/evaluators/, schemas/
+- Support short names: "contract-analyzer" → "schemas/agents/contract-analyzer.yaml"
+- Support relative/absolute paths
+- Consistent error messages and logging
+i
+Usage:
+    # From API
+    schema = load_agent_schema("rem")
+
+    # From CLI with custom path
+    schema = load_agent_schema("./my-agent.yaml")
+
+    # From agent factory
+    schema = load_agent_schema("contract-analyzer")
+
+Schema Caching Status:
+
+    ✅ IMPLEMENTED: Filesystem Schema Caching (2025-11-22)
+       - Schemas loaded from package resources cached indefinitely in _fs_schema_cache
+       - No TTL needed (immutable, versioned with code)
+       - Lazy-loaded on first access
+       - Custom paths not cached (may change during development)
+
+    TODO: Database Schema Caching (Future)
+       - Schemas loaded from schemas table (SchemaRepository)
+       - Will require TTL for cache invalidation (5-15 minutes)
+       - May change at runtime via admin updates
+       - Cache key: (schema_name, version) → (schema_dict, timestamp)
+       - Implementation ready in _db_schema_cache and _db_schema_ttl
+
+    Benefits Achieved:
+    - ✅ Eliminated disk I/O for repeated schema loads
+    - ✅ Faster agent creation (critical for API latency)
+    - 🔲 Database query reduction (pending DB schema implementation)
+
+    Future Enhancement (when database schemas are implemented):
+        import time
+
+        _db_schema_cache: dict[tuple[str, str], tuple[dict[str, Any], float]] = {}
+        _db_schema_ttl: int = 300  # 5 minutes
+
+        async def load_agent_schema_from_db(name: str, version: str | None = None):
+            cache_key = (name, version or "latest")
+            if cache_key in _db_schema_cache:
+                schema, timestamp = _db_schema_cache[cache_key]
+                if time.time() - timestamp < _db_schema_ttl:
+                    return schema
+            # Load from DB and cache with TTL
+            from rem.services.repositories import schema_repository
+            schema = await schema_repository.get_by_name(name, version)
+            _db_schema_cache[cache_key] = (schema, time.time())
+            return schema
+
+    Related:
+    - rem/src/rem/agentic/providers/pydantic_ai.py (create_agent factory)
+    - rem/src/rem/services/repositories/schema_repository.py (database schemas)
+"""
+
+import importlib.resources
+from pathlib import Path
+from typing import Any, cast
+
+import yaml
+from loguru import logger
+
+
+# Standard search paths for agent schemas (in priority order)
+SCHEMA_SEARCH_PATHS = [
+    "schemas/agents/{name}.yaml",          # Top-level agents (e.g., rem.yaml)
+    "schemas/agents/core/{name}.yaml",     # Core system agents
+    "schemas/agents/examples/{name}.yaml", # Example agents
+    "schemas/evaluators/{name}.yaml",
+    "schemas/{name}.yaml",
+]
+
+# In-memory cache for filesystem schemas (no TTL - immutable)
+_fs_schema_cache: dict[str, dict[str, Any]] = {}
+
+# Future: Database schema cache (with TTL - mutable)
+# Will be used when loading schemas from database (SchemaRepository)
+# _db_schema_cache: dict[tuple[str, str], tuple[dict[str, Any], float]] = {}
+# _db_schema_ttl: int = 300  # 5 minutes in seconds
+
+
+def load_agent_schema(schema_name_or_path: str, use_cache: bool = True) -> dict[str, Any]:
+    """
+    Load agent schema from YAML file with unified search logic and caching.
+
+    Filesystem schemas are cached indefinitely (immutable, versioned with code).
+    Database schemas (future) will be cached with TTL for invalidation.
+
+    Handles path resolution automatically:
+    - "rem" → searches schemas/agents/rem.yaml (top-level)
+    - "moment-builder" → searches schemas/agents/core/moment-builder.yaml
+    - "contract-analyzer" → searches schemas/agents/examples/contract-analyzer.yaml
+    - "core/moment-builder" → searches schemas/agents/core/moment-builder.yaml
+    - "/absolute/path.yaml" → loads directly
+    - "relative/path.yaml" → loads relative to cwd
+
+    Search Order:
+    1. Check cache (if use_cache=True and schema found in FS cache)
+    2. Exact path if it exists (absolute or relative)
+    3. Package resources: schemas/agents/{name}.yaml (top-level)
+    4. Package resources: schemas/agents/core/{name}.yaml
+    5. Package resources: schemas/agents/examples/{name}.yaml
+    6. Package resources: schemas/evaluators/{name}.yaml
+    7. Package resources: schemas/{name}.yaml
+
+    Args:
+        schema_name_or_path: Schema name or file path
+            Examples: "rem-query-agent", "contract-analyzer", "./my-schema.yaml"
+        use_cache: If True, uses in-memory cache for filesystem schemas
+
+    Returns:
+        Agent schema as dictionary
+
+    Raises:
+        FileNotFoundError: If schema not found in any search location
+        yaml.YAMLError: If schema file is invalid YAML
+
+    Examples:
+        >>> # Load by short name (cached after first load)
+        >>> schema = load_agent_schema("contract-analyzer")
+        >>>
+        >>> # Load from custom path (not cached - custom paths may change)
+        >>> schema = load_agent_schema("./my-agent.yaml")
+        >>>
+        >>> # Load evaluator schema (cached)
+        >>> schema = load_agent_schema("rem-lookup-correctness")
+    """
+    # Normalize the name for cache key
+    cache_key = str(schema_name_or_path).replace('agents/', '').replace('schemas/', '').replace('evaluators/', '').replace('core/', '').replace('examples/', '')
+    if cache_key.endswith('.yaml') or cache_key.endswith('.yml'):
+        cache_key = cache_key.rsplit('.', 1)[0]
+
+    # Check cache first (only for package resources, not custom paths)
+    path = Path(schema_name_or_path)
+    is_custom_path = path.exists() or '/' in str(schema_name_or_path) or '\\' in str(schema_name_or_path)
+
+    if use_cache and not is_custom_path and cache_key in _fs_schema_cache:
+        logger.debug(f"Loading schema from cache: {cache_key}")
+        return _fs_schema_cache[cache_key]
+
+    # 1. Try exact path first (absolute or relative to cwd)
+    if path.exists():
+        logger.debug(f"Loading schema from exact path: {path}")
+        with open(path, "r") as f:
+            schema = yaml.safe_load(f)
+        logger.debug(f"Loaded schema with keys: {list(schema.keys())}")
+        # Don't cache custom paths (they may change)
+        return cast(dict[str, Any], schema)
+
+    # 2. Normalize name for package resource search
+    base_name = cache_key
+
+    # 3. Try package resources with standard search paths
+    for search_pattern in SCHEMA_SEARCH_PATHS:
+        search_path = search_pattern.format(name=base_name)
+
+        try:
+            # Use importlib.resources to find schema in installed package
+            schema_ref = importlib.resources.files("rem") / search_path
+            schema_path = Path(str(schema_ref))
+
+            if schema_path.exists():
+                logger.debug(f"Loading schema from package: {search_path}")
+                with open(schema_path, "r") as f:
+                    schema = yaml.safe_load(f)
+                logger.debug(f"Loaded schema with keys: {list(schema.keys())}")
+
+                # Cache filesystem schemas (immutable, safe to cache indefinitely)
+                if use_cache:
+                    _fs_schema_cache[cache_key] = schema
+                    logger.debug(f"Cached schema: {cache_key}")
+
+                return cast(dict[str, Any], schema)
+        except Exception as e:
+            logger.debug(f"Could not load from {search_path}: {e}")
+            continue
+
+    # 4. Schema not found in any location
+    searched_paths = [pattern.format(name=base_name) for pattern in SCHEMA_SEARCH_PATHS]
+    raise FileNotFoundError(
+        f"Schema not found: {schema_name_or_path}\n"
+        f"Searched locations:\n"
+        f"  - Exact path: {path}\n"
+        f"  - Package resources: {', '.join(searched_paths)}"
+    )
+
+
+def validate_agent_schema(schema: dict[str, Any]) -> bool:
+    """
+    Validate agent schema structure.
+
+    Basic validation checks:
+    - Has 'type' field (should be 'object')
+    - Has 'description' field (system prompt)
+    - Has 'properties' field (output schema)
+
+    Args:
+        schema: Agent schema dict
+
+    Returns:
+        True if valid
+
+    Raises:
+        ValueError: If schema is invalid
+    """
+    if not isinstance(schema, dict):
+        raise ValueError(f"Schema must be a dict, got {type(schema)}")
+
+    if schema.get('type') != 'object':
+        raise ValueError(f"Schema type must be 'object', got {schema.get('type')}")
+
+    if 'description' not in schema:
+        raise ValueError("Schema must have 'description' field (system prompt)")
+
+    if 'properties' not in schema:
+        logger.warning("Schema missing 'properties' field - agent will have no structured output")
+
+    logger.debug("Schema validation passed")
+    return True