using-graphiti-memory

Using Graphiti Memory

When to Use

Query PROACTIVELY:

• ✅ Before implementing ANY feature
• ✅ When user mentions past work or patterns
• ✅ When architecting new components
• ✅ Before making technical decisions

Save Immediately After:

• ✅ Making architectural decisions
• ✅ Discovering reusable patterns
• ✅ Learning from bugs/issues
• ✅ Receiving user preferences

How to Query

Step 1: Get config and repo context (run once):

source "$HOME/.config/claude/graphiti-context-hub.conf" 2>/dev/null

GROUP_ID="${GRAPHITI_GROUP_ID:-main}"
REPO_NAME=$(git remote get-url origin 2>/dev/null | sed 's/.*\///' | sed 's/\.git$//' || basename "$PWD")

Step 2: Search Graphiti (use GROUP_ID from above):

# Search nodes (entities)
nodes_result = mcp__graphiti__search_nodes({
    "query": "authentication patterns",
    "group_ids": [GROUP_ID],  # Always "main"
    "max_nodes": 10
})

# Search facts (relationships)
facts_result = mcp__graphiti__search_memory_facts({
    "query": "authentication flow",
    "group_ids": [GROUP_ID],
    "max_facts": 20
})

How to Save

Step 1: Get config (if not already loaded):

source "$HOME/.config/claude/graphiti-context-hub.conf" 2>/dev/null

GROUP_ID="${GRAPHITI_GROUP_ID:-main}"
REPO_NAME=$(git remote get-url origin 2>/dev/null | sed 's/.*\///' | sed 's/\.git$//' || basename "$PWD")

Step 2: Save episode with repo tagging:

# IMPORTANT: Prefix episode_body with "Repo: {REPO_NAME}\n\n"
result = mcp__graphiti__add_memory({
    "name": "Auth Decision: JWT with httponly cookies",
    "episode_body": f"""Repo: {REPO_NAME}

Decision: Using JWT tokens stored in httponly cookies.

Rationale:
- XSS protection via httponly flag
- CSRF protection via SameSite attribute
- Automatic token rotation on refresh

Implementation:
- Access token: 15 min expiry
- Refresh token: 7 day expiry
- Redis for token storage
""",
    "group_id": GROUP_ID,  # Always "main"
    "source": "context-hub",
    "source_description": "Architectural decision"
})

episode_id = result.get('episode_id', result.get('uuid'))

Key Pattern: Always prefix episode_body with Repo: {REPO_NAME}\n\n to tag which repository the memory belongs to.

Graphiti Features

Automatic Entity Extraction:

• Graphiti extracts entities (JWT, Redis, tokens) automatically
• Creates relationships between entities
• No manual linking required

Search Capabilities:

• Semantic search across nodes (entities)
• Relationship search via facts
• Temporal queries (what changed when)

Knowledge Graph Benefits:

• See connections between decisions
• Trace architectural evolution
• Discover related patterns across projects

Configuration

Config file: ~/.config/claude/graphiti-context-hub.conf

GRAPHITI_GROUP_ID=main
GRAPHITI_ENDPOINT=http://localhost:8000
SERENA_ENABLED=true

Created by /context-hub-setup. Defaults: group_id=main, endpoint=localhost:8000.

Common Patterns

Before implementing a feature:

# Search for related patterns
nodes = mcp__graphiti__search_nodes({
    "query": f"implementing {feature_name}",
    "group_ids": [GROUP_ID],
    "max_nodes": 10
})

# Check for architectural decisions
facts = mcp__graphiti__search_memory_facts({
    "query": f"{feature_name} architecture",
    "group_ids": [GROUP_ID],
    "max_facts": 20
})

After making a decision:

mcp__graphiti__add_memory({
    "name": f"Decision: {decision_title}",
    "episode_body": f"""Repo: {REPO_NAME}

Decision: {what_you_decided}

Context: {why_this_matters}

Rationale: {reasoning}

Trade-offs: {alternatives_considered}
""",
    "group_id": GROUP_ID,
    "source": "context-hub"
})