Graphiti Temporal Context Skill

Inspired by: Graphiti — temporal context graphs for AI agents

Purpose: Integrate Graphiti's temporal knowledge graph patterns into local-llm-server's agent memory and context management.

What's Unique About Graphiti

Graphiti builds temporal context graphs (evolving knowledge graphs that track how facts change over time):

• Temporal awareness — knows what's true now vs what was true before
• Provenance tracking — maintains links to source data
• Hybrid retrieval — semantic + keyword + graph traversal
• Incremental updates — efficiently adds new information without full recomputation

Unlike traditional RAG (flat chunks), Graphiti gives agents rich, structured context that evolves with each interaction.

Integration Opportunities

1. Agent Memory as Temporal Graph

Track agent decisions and outcomes:

class AgentContextGraph:
    def add_interaction(self, timestamp, agent_id, action):
        """Record agent action with temporal metadata"""
    
    def query_at_time(self, entity, timestamp):
        """Query what was true at specific time"""

2. Multi-Agent Coordination

Track which agents worked on which tasks with temporal awareness.

3. Knowledge Queries

Query across relationships with SQL:

SELECT entity, fact, timestamp FROM context_graph
WHERE entity LIKE 'test_%'
  AND fact LIKE 'status:failed'
ORDER BY timestamp DESC;

Database Schema

CREATE TABLE temporal_context (
    id TEXT PRIMARY KEY,
    entity TEXT NOT NULL,
    fact TEXT NOT NULL,
    timestamp DATETIME NOT NULL,
    provenance TEXT,
    agent_id TEXT,
    created_at DATETIME DEFAULT CURRENT_TIMESTAMP
);

CREATE INDEX idx_entity_time ON temporal_context(entity, timestamp DESC);

Files to Create

• services/temporal_context.py — temporal graph implementation
• db/temporal_store.py — SQLite storage
• tests/test_temporal_context.py — tests

References

• Graphiti: https://github.com/getzep/graphiti
• Quick-Note Issue: #263