mem0-memory（内存 0 内存）

Mem0 Memory Management

Persist and retrieve semantic memories across Claude sessions.

Memory Scopes

Organize memories by scope for efficient retrieval:

Project Isolation

Memories are isolated by project name extracted from CLAUDE_PROJECT_DIR:

• Project name: basename($CLAUDE_PROJECT_DIR) (sanitized to lowercase, dashes)
• Format: {project-name}-{scope}

Edge Case: If two different repositories have the same directory name, they will share the same user_id scope. To avoid this:

1. Use unique directory names for each project
2. Or use MEM0_ORG_ID environment variable for additional namespace

Example:

• /Users/alice/my-app → my-app-decisions ✅
• /Users/bob/my-app → my-app-decisions ⚠️ (collision if same mem0.ai project)
• With MEM0_ORG_ID=acme: /Users/alice/my-app → acme-my-app-decisions ✅

Memory Categories

Memories are automatically categorized based on content. Available categories:

Cross-Tool Memory

Memories include source_tool metadata to support cross-tool memory sharing:

• source_tool: "orchestkit-claude" - Memories from Claude Code
• source_tool: "orchestkit-cursor" - Memories from Cursor (future)

Query memories by tool:

# Query Claude Code memories
filters={"AND": [{"metadata.source_tool": "orchestkit-claude"}]}

# Query all memories (any tool)
filters={"AND": [{"user_id": "my-project-decisions"}]}

Setup

Install mem0 Python SDK:

# Install the mem0ai package and dependencies
pip install mem0ai python-dotenv

# Or install from requirements file
pip install -r skills/mem0-memory/scripts/requirements.txt

Optional - Install mem0-skill-lib package (recommended for development):

# Install as editable package for proper type checking
pip install -e skills/mem0-memory/scripts/

Note: Scripts work in both modes:

• Standalone mode (default): Scripts dynamically add lib/ to sys.path. Type checkers require # type: ignore comments for these dynamic imports.
• Installed mode: If mem0-skill-lib is installed, scripts import from the installed package without type ignore comments.

Set environment variables:

Option 1: Using .env file (Recommended)

Create a .env file in your project root:

# Copy the example file
cp .env.example .env

# Edit .env and add your API key
MEM0_API_KEY=sk-your-api-key-here
MEM0_ORG_ID=org_...      # Optional (for organization-level scoping)
MEM0_PROJECT_ID=proj_... # Optional (Pro feature)
MEM0_WEBHOOK_URL=https://your-domain.com/webhook/mem0  # Optional

The scripts automatically load from .env if it exists.

Option 2: Shell environment variables

export MEM0_API_KEY="sk-..."
export MEM0_ORG_ID="org_..."      # Optional (for organization-level scoping)
export MEM0_PROJECT_ID="proj_..." # Optional (Pro feature)

Verify installation:

python3 -c "from mem0 import MemoryClient; print('✓ mem0ai installed successfully')"

Core Operations

Adding Memories

Execute the script via Bash tool:

!bash skills/mem0-memory/scripts/crud/add-memory.py \
  --text "Decided to use FastAPI over Flask for async support" \
  --user-id "project-decisions" \
  --metadata '{"scope":"project-decisions","category":"backend","date":"2026-01-12"}' \
  --enable-graph

Best practices for adding:

• Be specific and actionable
• Include rationale ("because...")
• Add scope and category metadata
• Timestamp important decisions

Searching Memories

!bash skills/mem0-memory/scripts/crud/search-memories.py \
  --query "authentication approach" \
  --user-id "project-decisions" \
  --limit 5 \
  --enable-graph

Search tips:

• Use natural language queries
• Search by topic, not exact phrases
• Combine with scope filters when available
• Enable graph (--enable-graph) to get relationship information in results

Graph Relationships in Search Results:

When --enable-graph is enabled, search results include:

• relations array with relationship information
• related_via field showing how results are connected
• relationship_summary with relation types found

Graph Relationship Queries

Get Related Memories:

Query memories related to a given memory via graph traversal:

!bash skills/mem0-memory/scripts/graph/get-related-memories.py \
  --memory-id "mem_abc123" \
  --depth 2 \
  --relation-type "recommends"

Traverse Graph:

Multi-hop graph traversal for complex relationship queries:

!bash skills/mem0-memory/scripts/graph/traverse-graph.py \
  --memory-id "mem_abc123" \
  --depth 2 \
  --relation-type "recommends"

Example Use Cases:

1. Multi-hop queries: "What did database-engineer recommend about pagination?" → Traverses: database-engineer → recommends → cursor-pagination → Returns related memories with relationship context
2. Context expansion: Find a memory about "authentication" → Get related memories via graph (depth 2) → Discover related decisions, patterns, and recommendations
3. Relationship filtering: --relation-type "recommends" # Only follow "recommends" relationships --relation-type "uses" # Only follow "uses" relationships

Listing Memories

!bash skills/mem0-memory/scripts/crud/get-memories.py \
  --user-id "project-orchestkit" \
  --filters '{"limit":100}'

Getting Single Memory

!bash skills/mem0-memory/scripts/crud/get-memory.py \
  --memory-id "mem_abc123"

Updating Memories

!bash skills/mem0-memory/scripts/crud/update-memory.py \
  --memory-id "mem_abc123" \
  --text "Updated decision text" \
  --metadata '{"updated":true}'

Deleting Memories

!bash skills/mem0-memory/scripts/crud/delete-memory.py \
  --memory-id "mem_abc123"

When to delete:

• Outdated decisions that were reversed
• Incorrect information
• Duplicate or redundant entries

What to Remember

Good candidates:

• Architecture decisions with rationale
• API contracts and interfaces
• Naming conventions adopted
• Technical debt acknowledged
• Blockers and their resolutions
• User preferences and style

Avoid storing:

• Temporary debugging context
• Large code blocks (use Git)
• Secrets or credentials
• Highly volatile information

Memory Patterns

Decision Memory

"Decision: Use cursor-based pagination for all list endpoints.
Rationale: Better performance for large datasets, consistent UX.
Date: 2026-01-12. Scope: API design."

Pattern Memory

"Pattern: All React components export default function.
Convention: Use named exports only for utilities.
Applies to: frontend/src/components/**"

Continuity Memory

"Session handoff: Completed hybrid search implementation.
Next steps: Add metadata boosting, write integration tests.
PR #456 ready for review. Blocked on: DB migration approval."

Advanced Operations (Pro Features)

Batch Operations

Batch update up to 1000 memories:

!bash skills/mem0-memory/scripts/batch/batch-update.py \
  --memories '[{"memory_id":"mem_123","text":"updated"},{"memory_id":"mem_456","metadata":{"updated":true}}]'

Batch delete:

!bash skills/mem0-memory/scripts/batch/batch-delete.py \
  --memory-ids '["mem_123","mem_456","mem_789"]'

Memory History (Audit Trail)

!bash skills/mem0-memory/scripts/utils/memory-history.py \
  --memory-id "mem_abc123"

Exports (Data Portability)

Create export:

!bash skills/mem0-memory/scripts/export/export-memories.py \
  --filters '{"user_id":"project-decisions"}' \
  --schema '{"format":"json"}'

Retrieve export:

!bash skills/mem0-memory/scripts/export/get-export.py \
  --user-id "project-decisions"

Analytics

Get memory statistics:

!bash skills/mem0-memory/scripts/utils/memory-summary.py \
  --filters '{"user_id":"project-decisions"}'

List all users:

!bash skills/mem0-memory/scripts/utils/get-users.py

Webhooks (Automation)

!bash skills/mem0-memory/scripts/webhooks/create-webhook.py \
  --url "https://example.com/webhook" \
  --name "Memory Webhook" \
  --event-types '["memory.created","memory.updated"]'

Integration with OrchestKit

Use memories to maintain context across plugin sessions:

# At session start - recall project context
!bash skills/mem0-memory/scripts/crud/search-memories.py \
  --query "current sprint priorities" \
  --user-id "project-continuity"

# During work - persist decisions
!bash skills/mem0-memory/scripts/crud/add-memory.py \
  --text "Implemented feature using approach because reason" \
  --user-id "project-decisions" \
  --metadata '{"scope":"project-decisions"}'

# At session end - save continuity
!bash skills/mem0-memory/scripts/crud/add-memory.py \
  --text "Session end: summary. Next: next_steps" \
  --user-id "project-continuity" \
  --metadata '{"scope":"project-continuity"}'

Scripts Available

All scripts are located in skills/mem0-memory/scripts/:

Core Scripts:

• add-memory.py - Store new memory
• search-memories.py - Semantic search
• get-memories.py - List all memories (with filters)
• get-memory.py - Get single memory by ID
• update-memory.py - Update memory content/metadata
• delete-memory.py - Remove memory

Advanced Scripts (Pro Features):

• batch-update.py - Bulk update up to 1000 memories
• batch-delete.py - Bulk delete multiple memories
• memory-history.py - Get audit trail for a memory
• export-memories.py - Create structured export
• get-export.py - Retrieve export data
• memory-summary.py - Get statistics/analytics
• get-events.py - Track async operations
• get-users.py - List all users (analytics)
• create-webhook.py - Setup webhooks for automation

Note: MCP integration is deprecated. Use scripts instead for better control, versioning, and access to all 30+ API methods.

Related Skills

• semantic-caching - Semantic caching patterns that complement long-term memory
• embeddings - Embedding strategies used by Mem0 for semantic search
• langgraph-checkpoints - State persistence patterns for workflow continuity
• context-compression - Compress context when memory retrieval adds too many tokens

Key Decisions

Capability Details

memory-add

Keywords: add memory, remember, store, persist, save context Solves:

• How do I save information for later sessions?
• Persist a decision or pattern
• Store project context across sessions

memory-search

Keywords: search memory, recall, find, retrieve, what did we Solves:

• How do I find previous decisions?
• Recall context from past sessions
• Search for specific patterns or conventions

memory-list

Keywords: list memories, show all, get memories, view stored Solves:

• How do I see all stored memories?
• List project decisions
• Review stored patterns

memory-delete

Keywords: delete memory, forget, remove, clear Solves:

• How do I remove outdated memories?
• Delete incorrect information
• Clean up duplicate entries