Memory Observer

Purpose

The Memory Observer ensures the system is not a "black box". It provides visibility into how memory is being accessed, modified, and evolved over time.

Key Features

• Real-time Streams: Pushes access events to Redis streams for live dashboards.
• Graph Visualization: Exports GraphML snapshots for UI rendering of memory structures.

Functional Deep Dive

Logging

Logs every memory access to SQL memory_access_logs and updates memory_ttl timestamps to delay pruning for active items.

Metrics

Updates Prometheus counters (access frequency, latency) and gauges (total item counts, storage savings from deduplication).

Streaming

Pushes real-time access events to Redis streams, allowing admin dashboards to show a live "pulse" of the memory system.

API Reference

Memory Observer provides endpoints for monitoring and metrics.

SDK Reference

The Functor SDK provides a Python interface for observability operations.

Monitoring and Metrics

from functor_sdk import FunctorClient

client = FunctorClient(api_key="your-api-key")

# Log memory access
client.memory.observability.log_access(
    item_id="item_123",
    item_type="episode",
    access_type="read",
    user_id="user_456",
    session_id="session_789"
)

# Get memory statistics
stats = client.memory.observability.get_stats(
    kg_name="user_knowledge",
    time_window_hours=24
)

# Query access logs
logs = client.memory.observability.get_access_logs(
    user_id="user_456",
    item_type="episode",
    limit=100
)

# Get Prometheus metrics
metrics = client.memory.observability.get_metrics()

# Get module health
health = client.memory.observability.get_module_health(module="episodic")

# Visualize knowledge graph
viz = client.memory.observability.visualize(
    kg_name="user_knowledge",
    format="json"
)