### Step 2: Define Your Project's Memory Schema

Add this section to your `CLAUDE.md` to define project-specific entities:

```markdown
## Project-Specific Entities

This project tracks the following custom entities:

### Entity: Component
Represents a software component in our architecture.

Extract when: User mentions services, modules, libraries, or system parts
```python
{
    "type": "Component",
    "name": "AuthService",
    "component_type": "service",
    "description": "Handles user authentication",
    "dependencies": ["Database", "CacheService"],
    "owner": "backend-team"
}

### Step 3: Create Project Entity Definitions

Create `ai/graph/entities/project_entities.py` in your project:

```python
"""
Project-specific entities for the knowledge graph.
Copy this file to your project at: ai/graph/entities/project_entities.py
"""

from pydantic import BaseModel, Field
from typing import List, Optional

class Component(BaseModel):
    """
    Software component in the system architecture.
  
    Extract when user mentions:
    - Services, microservices
    - Modules, packages
    - Libraries, frameworks
    - Databases, caches
    - Any architectural component
    """
    name: str = Field(description="Component name")
    component_type: str = Field(description="Type: service|library|database|cache|module")
    description: str = Field(description="What this component does")
    dependencies: List[str] = Field(description="Other components this depends on")
    owner: Optional[str] = Field(description="Team or person responsible")
  
class Feature(BaseModel):
    """
    Product feature or user story.
  
    Extract when user mentions:
    - Features, functionality
    - User stories, requirements
    - Capabilities, use cases
    """
    name: str = Field(description="Feature name")
    status: str = Field(description="Status: planned|in_progress|completed|blocked")
    priority: str = Field(description="Priority: critical|high|medium|low")
    description: str = Field(description="What this feature does")
    assigned_to: Optional[str] = Field(description="Developer assigned")
    related_components: List[str] = Field(description="Components involved")

class Configuration(BaseModel):
    """
    System configuration or settings.
  
    Extract when user mentions:
    - Config, configuration
    - Settings, parameters
    - Environment variables
    - System properties
    """
    name: str = Field(description="Configuration name")
    environment: str = Field(description="Environment: dev|staging|prod|local")
    category: str = Field(description="Category: database|api|security|performance")
    settings: dict = Field(description="Key-value configuration pairs")

class TechnicalDecision(BaseModel):
    """
    Architectural or implementation decision.
  
    Extract when user:
    - Makes technology choices
    - Decides between alternatives
    - Establishes patterns
    - Sets technical direction
    """
    decision_type: str = Field(description="Type: architecture|tool|pattern|approach")
    chosen: str = Field(description="What was chosen")
    alternatives: List[str] = Field(description="Options that were considered")
    rationale: str = Field(description="Why this choice was made")
    impact: Optional[str] = Field(description="Expected impact or trade-offs")
    date: str = Field(description="When decision was made")

class Issue(BaseModel):
    """
    Bug, problem, or issue encountered.
  
    Extract when user mentions:
    - Bugs, errors, issues
    - Problems, failures
    - Things not working
    - Unexpected behavior
    """
    title: str = Field(description="Brief issue description")
    severity: str = Field(description="Severity: critical|high|medium|low")
    status: str = Field(description="Status: open|investigating|resolved")
    component: str = Field(description="Affected component")
    description: str = Field(description="Detailed problem description")
    solution: Optional[str] = Field(description="How it was resolved")
    root_cause: Optional[str] = Field(description="Why it happened")

### What This Gives You

With these files in place, Claude will:

1. **Always search before acting** - No more repeating mistakes or ignoring preferences
2. **Build memory throughout conversations** - Every interaction adds to the knowledge
3. **Remember across sessions** - Previous decisions and patterns persist
4. **Understand your project deeply** - Accumulates context over time
5. **Follow established patterns** - Respects preferences and procedures

The key is that the instructions make memory usage MANDATORY, not optional. Claude must search before every action and capture information continuously.

## Danger zone
Setting `NEO4J_DESTROY_ENTIRE_GRAPH=true` wipes *all* projects the next time you run `graphiti up`. Use with care.

## Contributing
PRs and issues are welcome.

© 2025 rawr‑ai • MIT License