alma-memory 0.5.0__py3-none-any.whl → 0.5.1__py3-none-any.whl

alma/graph/backends/memgraph.py

@@ -0,0 +1,432 @@
+"""
+ALMA Graph Memory - Memgraph Backend.
+
+Memgraph implementation of the GraphBackend interface.
+Memgraph is compatible with Neo4j's Bolt protocol, so the neo4j Python driver works with it.
+"""
+
+import json
+import logging
+from datetime import datetime, timezone
+from typing import Any, Dict, List, Optional
+
+from alma.graph.base import GraphBackend
+from alma.graph.store import Entity, Relationship
+
+logger = logging.getLogger(__name__)
+
+
+class MemgraphBackend(GraphBackend):
+    """
+    Memgraph graph database backend.
+
+    Memgraph is an in-memory graph database compatible with Neo4j's Bolt protocol.
+    This backend uses the neo4j Python driver.
+
+    Requires neo4j Python driver: pip install neo4j
+
+    Example usage:
+        backend = MemgraphBackend(
+            uri="bolt://localhost:7687",
+            username="",
+            password=""
+        )
+        backend.add_entity(entity)
+        backend.close()
+
+    Note: Memgraph typically doesn't require authentication by default,
+    but it can be configured. Use empty strings for username/password if
+    authentication is disabled.
+    """
+
+    def __init__(
+        self,
+        uri: str = "bolt://localhost:7687",
+        username: str = "",
+        password: str = "",
+        database: str = "memgraph",
+    ):
+        """
+        Initialize Memgraph connection.
+
+        Args:
+            uri: Memgraph connection URI (bolt://)
+            username: Database username (empty if auth disabled)
+            password: Database password (empty if auth disabled)
+            database: Database name (default: "memgraph")
+        """
+        self.uri = uri
+        self.username = username
+        self.password = password
+        self.database = database
+        self._driver = None
+
+    def _get_driver(self):
+        """Lazy initialization of Memgraph driver (via neo4j package)."""
+        if self._driver is None:
+            try:
+                from neo4j import GraphDatabase
+
+                # Memgraph uses the same Bolt protocol as Neo4j
+                if self.username and self.password:
+                    self._driver = GraphDatabase.driver(
+                        self.uri,
+                        auth=(self.username, self.password),
+                    )
+                else:
+                    # No authentication
+                    self._driver = GraphDatabase.driver(self.uri)
+            except ImportError as err:
+                raise ImportError(
+                    "neo4j package required for Memgraph graph backend. "
+                    "Install with: pip install neo4j"
+                ) from err
+        return self._driver
+
+    def _run_query(self, query: str, parameters: Optional[Dict] = None) -> List[Dict]:
+        """Execute a Cypher query."""
+        driver = self._get_driver()
+        # Memgraph doesn't use the database parameter in the same way as Neo4j
+        # Most Memgraph setups use a single database
+        with driver.session() as session:
+            result = session.run(query, parameters or {})
+            return [dict(record) for record in result]
+
+    def add_entity(self, entity: Entity) -> str:
+        """Add or update an entity in Memgraph."""
+        # Extract project_id and agent from properties if present
+        properties = entity.properties.copy()
+        project_id = properties.pop("project_id", None)
+        agent = properties.pop("agent", None)
+
+        query = """
+        MERGE (e:Entity {id: $id})
+        SET e.name = $name,
+            e.entity_type = $entity_type,
+            e.properties = $properties,
+            e.created_at = $created_at
+        """
+        params = {
+            "id": entity.id,
+            "name": entity.name,
+            "entity_type": entity.entity_type,
+            "properties": json.dumps(properties),
+            "created_at": entity.created_at.isoformat(),
+        }
+
+        # Add optional fields if present
+        if project_id:
+            query += ", e.project_id = $project_id"
+            params["project_id"] = project_id
+        if agent:
+            query += ", e.agent = $agent"
+            params["agent"] = agent
+
+        query += " RETURN e.id as id"
+
+        result = self._run_query(query, params)
+        return result[0]["id"] if result else entity.id
+
+    def add_relationship(self, relationship: Relationship) -> str:
+        """Add or update a relationship in Memgraph."""
+        # Sanitize relationship type for Cypher (remove special characters)
+        rel_type = (
+            relationship.relation_type.replace("-", "_").replace(" ", "_").upper()
+        )
+
+        query = f"""
+        MATCH (source:Entity {{id: $source_id}})
+        MATCH (target:Entity {{id: $target_id}})
+        MERGE (source)-[r:{rel_type}]->(target)
+        SET r.id = $id,
+            r.properties = $properties,
+            r.confidence = $confidence,
+            r.created_at = $created_at
+        RETURN r.id as id
+        """
+        result = self._run_query(
+            query,
+            {
+                "id": relationship.id,
+                "source_id": relationship.source_id,
+                "target_id": relationship.target_id,
+                "properties": json.dumps(relationship.properties),
+                "confidence": relationship.confidence,
+                "created_at": relationship.created_at.isoformat(),
+            },
+        )
+        return result[0]["id"] if result else relationship.id
+
+    def get_entity(self, entity_id: str) -> Optional[Entity]:
+        """Get an entity by ID."""
+        query = """
+        MATCH (e:Entity {id: $id})
+        RETURN e.id as id, e.name as name, e.entity_type as entity_type,
+               e.properties as properties, e.created_at as created_at,
+               e.project_id as project_id, e.agent as agent
+        """
+        result = self._run_query(query, {"id": entity_id})
+        if not result:
+            return None
+
+        r = result[0]
+        properties = json.loads(r["properties"]) if r["properties"] else {}
+
+        # Add project_id and agent back to properties if present
+        if r.get("project_id"):
+            properties["project_id"] = r["project_id"]
+        if r.get("agent"):
+            properties["agent"] = r["agent"]
+
+        return Entity(
+            id=r["id"],
+            name=r["name"],
+            entity_type=r["entity_type"],
+            properties=properties,
+            created_at=(
+                datetime.fromisoformat(r["created_at"])
+                if r["created_at"]
+                else datetime.now(timezone.utc)
+            ),
+        )
+
+    def get_entities(
+        self,
+        entity_type: Optional[str] = None,
+        project_id: Optional[str] = None,
+        agent: Optional[str] = None,
+        limit: int = 100,
+    ) -> List[Entity]:
+        """Get entities with optional filtering."""
+        conditions = []
+        params: Dict[str, Any] = {"limit": limit}
+
+        if entity_type:
+            conditions.append("e.entity_type = $entity_type")
+            params["entity_type"] = entity_type
+        if project_id:
+            conditions.append("e.project_id = $project_id")
+            params["project_id"] = project_id
+        if agent:
+            conditions.append("e.agent = $agent")
+            params["agent"] = agent
+
+        where_clause = f"WHERE {' AND '.join(conditions)}" if conditions else ""
+
+        query = f"""
+        MATCH (e:Entity)
+        {where_clause}
+        RETURN e.id as id, e.name as name, e.entity_type as entity_type,
+               e.properties as properties, e.created_at as created_at,
+               e.project_id as project_id, e.agent as agent
+        LIMIT $limit
+        """
+
+        results = self._run_query(query, params)
+        entities = []
+        for r in results:
+            properties = json.loads(r["properties"]) if r["properties"] else {}
+            if r.get("project_id"):
+                properties["project_id"] = r["project_id"]
+            if r.get("agent"):
+                properties["agent"] = r["agent"]
+
+            entities.append(
+                Entity(
+                    id=r["id"],
+                    name=r["name"],
+                    entity_type=r["entity_type"],
+                    properties=properties,
+                    created_at=(
+                        datetime.fromisoformat(r["created_at"])
+                        if r["created_at"]
+                        else datetime.now(timezone.utc)
+                    ),
+                )
+            )
+        return entities
+
+    def get_relationships(self, entity_id: str) -> List[Relationship]:
+        """Get all relationships for an entity (both directions)."""
+        query = """
+        MATCH (e:Entity {id: $entity_id})-[r]-(other:Entity)
+        RETURN r.id as id,
+               CASE WHEN startNode(r).id = $entity_id THEN e.id ELSE other.id END as source_id,
+               CASE WHEN endNode(r).id = $entity_id THEN e.id ELSE other.id END as target_id,
+               type(r) as relation_type, r.properties as properties,
+               r.confidence as confidence, r.created_at as created_at
+        """
+
+        results = self._run_query(query, {"entity_id": entity_id})
+        relationships = []
+        for r in results:
+            rel_id = (
+                r["id"] or f"{r['source_id']}-{r['relation_type']}-{r['target_id']}"
+            )
+            relationships.append(
+                Relationship(
+                    id=rel_id,
+                    source_id=r["source_id"],
+                    target_id=r["target_id"],
+                    relation_type=r["relation_type"],
+                    properties=json.loads(r["properties"]) if r["properties"] else {},
+                    confidence=r["confidence"] or 1.0,
+                    created_at=(
+                        datetime.fromisoformat(r["created_at"])
+                        if r["created_at"]
+                        else datetime.now(timezone.utc)
+                    ),
+                )
+            )
+        return relationships
+
+    def search_entities(
+        self,
+        query: str,
+        embedding: Optional[List[float]] = None,
+        top_k: int = 10,
+    ) -> List[Entity]:
+        """
+        Search for entities by name.
+
+        Note: Vector similarity search requires Memgraph MAGE with vector operations.
+        Falls back to text search if embedding is provided but vector index
+        is not available.
+        """
+        # For now, we do text-based search
+        # Vector search can be added when Memgraph MAGE is configured
+        cypher = """
+        MATCH (e:Entity)
+        WHERE toLower(e.name) CONTAINS toLower($query)
+        RETURN e.id as id, e.name as name, e.entity_type as entity_type,
+               e.properties as properties, e.created_at as created_at,
+               e.project_id as project_id, e.agent as agent
+        LIMIT $limit
+        """
+
+        results = self._run_query(cypher, {"query": query, "limit": top_k})
+        entities = []
+        for r in results:
+            properties = json.loads(r["properties"]) if r["properties"] else {}
+            if r.get("project_id"):
+                properties["project_id"] = r["project_id"]
+            if r.get("agent"):
+                properties["agent"] = r["agent"]
+
+            entities.append(
+                Entity(
+                    id=r["id"],
+                    name=r["name"],
+                    entity_type=r["entity_type"],
+                    properties=properties,
+                    created_at=(
+                        datetime.fromisoformat(r["created_at"])
+                        if r["created_at"]
+                        else datetime.now(timezone.utc)
+                    ),
+                )
+            )
+        return entities
+
+    def delete_entity(self, entity_id: str) -> bool:
+        """Delete an entity and its relationships."""
+        query = """
+        MATCH (e:Entity {id: $id})
+        DETACH DELETE e
+        RETURN count(e) as deleted
+        """
+        result = self._run_query(query, {"id": entity_id})
+        return result[0]["deleted"] > 0 if result else False
+
+    def delete_relationship(self, relationship_id: str) -> bool:
+        """Delete a specific relationship by ID."""
+        query = """
+        MATCH ()-[r]-()
+        WHERE r.id = $id
+        DELETE r
+        RETURN count(r) as deleted
+        """
+        result = self._run_query(query, {"id": relationship_id})
+        return result[0]["deleted"] > 0 if result else False
+
+    def close(self) -> None:
+        """Close the Memgraph driver connection."""
+        if self._driver:
+            self._driver.close()
+            self._driver = None
+
+    # Additional methods for compatibility with existing GraphStore API
+
+    def find_entities(
+        self,
+        name: Optional[str] = None,
+        entity_type: Optional[str] = None,
+        limit: int = 10,
+    ) -> List[Entity]:
+        """
+        Find entities by name or type.
+
+        This method provides compatibility with the existing GraphStore API.
+        """
+        if name:
+            return self.search_entities(query=name, top_k=limit)
+
+        return self.get_entities(entity_type=entity_type, limit=limit)
+
+    def get_relationships_directional(
+        self,
+        entity_id: str,
+        direction: str = "both",
+        relation_type: Optional[str] = None,
+    ) -> List[Relationship]:
+        """
+        Get relationships for an entity with direction control.
+
+        This method provides compatibility with the existing GraphStore API.
+
+        Args:
+            entity_id: The entity ID.
+            direction: "outgoing", "incoming", or "both".
+            relation_type: Optional filter by relationship type.
+
+        Returns:
+            List of matching relationships.
+        """
+        if direction == "outgoing":
+            pattern = "(e)-[r]->(other)"
+        elif direction == "incoming":
+            pattern = "(e)<-[r]-(other)"
+        else:
+            pattern = "(e)-[r]-(other)"
+
+        type_filter = f":{relation_type}" if relation_type else ""
+
+        query = f"""
+        MATCH (e:Entity {{id: $entity_id}}){pattern.replace("[r]", f"[r{type_filter}]")}
+        RETURN r.id as id, e.id as source_id, other.id as target_id,
+               type(r) as relation_type, r.properties as properties,
+               r.confidence as confidence, r.created_at as created_at
+        """
+
+        results = self._run_query(query, {"entity_id": entity_id})
+        relationships = []
+        for r in results:
+            rel_id = (
+                r["id"] or f"{r['source_id']}-{r['relation_type']}-{r['target_id']}"
+            )
+            relationships.append(
+                Relationship(
+                    id=rel_id,
+                    source_id=r["source_id"],
+                    target_id=r["target_id"],
+                    relation_type=r["relation_type"],
+                    properties=json.loads(r["properties"]) if r["properties"] else {},
+                    confidence=r["confidence"] or 1.0,
+                    created_at=(
+                        datetime.fromisoformat(r["created_at"])
+                        if r["created_at"]
+                        else datetime.now(timezone.utc)
+                    ),
+                )
+            )
+        return relationships

alma/integration/claude_agents.py

@@ -212,8 +212,8 @@ class ClaudeAgentHooks:
             self.harness.post_run(harness_context, run_result)
             return True
         else:
-            # Direct ALMA learning
-            return self.alma.learn(
+            # Direct ALMA learning - now returns Outcome object
+            self.alma.learn(
                 agent=self.agent_name,
                 task=context.task_description,
                 outcome="success" if outcome.success else "failure",
@@ -223,6 +223,7 @@ class ClaudeAgentHooks:
                 error_message=outcome.error_message,
                 feedback=outcome.feedback,
             )
+            return True
 
     def format_memories_for_prompt(
         self,
@@ -312,15 +313,26 @@ class ClaudeAgentHooks:
             source: How this was discovered
 
         Returns:
-            True if knowledge was added, False if scope violation
+            True if knowledge was added
+
+        Raises:
+            ScopeViolationError: If domain is not within agent's scope
         """
-        result = self.alma.add_domain_knowledge(
-            agent=self.agent_name,
-            domain=domain,
-            fact=fact,
-            source=source,
-        )
-        return result is not None
+        from alma.exceptions import ScopeViolationError
+
+        try:
+            self.alma.add_domain_knowledge(
+                agent=self.agent_name,
+                domain=domain,
+                fact=fact,
+                source=source,
+            )
+            return True
+        except ScopeViolationError:
+            logger.warning(
+                f"[{self.agent_name}] Scope violation: cannot add knowledge in domain '{domain}'"
+            )
+            return False
 
 
 class AgentIntegration:

alma/learning/protocols.py

@@ -67,7 +67,7 @@ class LearningProtocol:
         duration_ms: Optional[int] = None,
         error_message: Optional[str] = None,
         feedback: Optional[str] = None,
-    ) -> bool:
+    ) -> Outcome:
         """
         Learn from a task outcome.
 
@@ -85,7 +85,7 @@ class LearningProtocol:
             feedback: User feedback
 
         Returns:
-            True if learning was accepted
+            The created Outcome record
         """
         # Validate agent has a scope (warn but don't block)
         scope = self.scopes.get(agent)
@@ -133,7 +133,7 @@ class LearningProtocol:
                 error=error_message,
             )
 
-        return True
+        return outcome_record
 
     def _maybe_create_heuristic(
         self,

alma/mcp/tools.py

@@ -172,7 +172,7 @@ def alma_learn(
         return {"success": False, "error": "strategy_used cannot be empty"}
 
     try:
-        result = alma.learn(
+        outcome_record = alma.learn(
             agent=agent,
             task=task,
             outcome=outcome,
@@ -185,10 +185,14 @@ def alma_learn(
 
         return {
             "success": True,
-            "learned": result,
-            "message": (
-                "Outcome recorded" if result else "Learning rejected (scope violation)"
-            ),
+            "learned": True,
+            "outcome": {
+                "id": outcome_record.id,
+                "agent": outcome_record.agent,
+                "task_type": outcome_record.task_type,
+                "success": outcome_record.success,
+            },
+            "message": "Outcome recorded successfully",
         }
 
     except Exception as e:
@@ -290,12 +294,6 @@ def alma_add_knowledge(
             source=source,
         )
 
-        if knowledge is None:
-            return {
-                "success": False,
-                "error": f"Agent '{agent}' not allowed to learn in domain '{domain}'",
-            }
-
         return {
             "success": True,
             "knowledge": {

alma/observability/__init__.py

@@ -0,0 +1,84 @@
+"""
+ALMA Observability Module.
+
+Provides comprehensive observability features including:
+- OpenTelemetry integration for distributed tracing
+- Structured JSON logging
+- Metrics collection (counters, histograms, gauges)
+- Performance monitoring
+
+This module follows the OpenTelemetry specification and supports
+integration with common observability backends (Jaeger, Prometheus,
+DataDog, etc.).
+
+Usage:
+    from alma.observability import (
+        get_tracer,
+        get_meter,
+        get_logger,
+        configure_observability,
+        ALMAMetrics,
+    )
+
+    # Initialize observability (typically at app startup)
+    configure_observability(
+        service_name="alma-memory",
+        enable_tracing=True,
+        enable_metrics=True,
+        log_format="json",
+    )
+
+    # Use in code
+    tracer = get_tracer(__name__)
+    with tracer.start_as_current_span("my_operation"):
+        # ... your code
+        pass
+"""
+
+from alma.observability.config import (
+    ObservabilityConfig,
+    configure_observability,
+    shutdown_observability,
+)
+from alma.observability.logging import (
+    JSONFormatter,
+    StructuredLogger,
+    get_logger,
+    setup_logging,
+)
+from alma.observability.metrics import (
+    ALMAMetrics,
+    MetricsCollector,
+    get_meter,
+    get_metrics,
+)
+from alma.observability.tracing import (
+    SpanKind,
+    TracingContext,
+    get_tracer,
+    trace_async,
+    trace_method,
+)
+
+__all__ = [
+    # Configuration
+    "ObservabilityConfig",
+    "configure_observability",
+    "shutdown_observability",
+    # Logging
+    "JSONFormatter",
+    "StructuredLogger",
+    "get_logger",
+    "setup_logging",
+    # Metrics
+    "ALMAMetrics",
+    "MetricsCollector",
+    "get_meter",
+    "get_metrics",
+    # Tracing
+    "SpanKind",
+    "TracingContext",
+    "get_tracer",
+    "trace_method",
+    "trace_async",
+]