AI Agent 知识图谱记忆设计提示词

You are a knowledge graph memory architect. Design a temporal context graph schema for an AI agent that needs to maintain persistent, evolving memory.

Agent Context

• Agent Purpose: [describe the agent's primary function]
• Data Sources: [list the types of information the agent processes]
• Update Frequency: [real-time / hourly / daily]
• Query Patterns: [what questions will the agent need to answer from memory?]

Design Requirements

1. Entity Schema

Define the core entity types with their properties:

Entity Type | Key Properties | Update Strategy
------------|---------------|----------------
[e.g. User] | [name, preferences, ...] | [merge/overwrite/append]

2. Relationship Types

Define how entities connect:

Relationship | From > To | Temporal? | Example
-------------|-----------|-----------|--------
[e.g. PREFERS] | User > Product | Yes | "Alice prefers dark roast (since 2024-01)"

3. Temporal Tracking Rules

• How to handle conflicting facts
• When to create a new version vs. update existing
• How to mark facts as superseded vs. deleted
• Validity window format: [valid_from, valid_to, confidence]

4. Retrieval Strategy

Design hybrid retrieval combining:

• Semantic search: for fuzzy concept matching
• Keyword search: for exact entity/fact lookup
• Graph traversal: for relationship-based queries
• Temporal filter: for "as of date X" queries

5. Ingestion Pipeline

For each new piece of information:

1. Extract entities and relationships
2. Resolve against existing graph (dedup, merge)
3. Detect contradictions with existing facts
4. Update validity windows
5. Maintain provenance (source episode > fact)

Output the complete schema design with examples for each component.