Deep Memory Search

Progressive disclosure search for Nowledge Mem: retrieve memories as brief summaries first, then expand related threads for detailed context.

When to Use

This skill handles requests involving:

• Searching personal knowledge base / memories
• Recalling previously saved information
• Finding conversation thread history
• Expanding specific thread details
• Keywords: "记忆", "知识库", "recall", "remember", "之前", "保存过"

Workflow

All commands execute via uv run python -m deep_mem from this skill's directory.

Step 1: Search Memories

uv run python -m deep_mem search "<user_query>"

Options: | Flag | Description | |------|-------------| | --limit N | Max memories to return (default: 10) | | --verbose | Show longer content previews | | --no-threads | Skip thread discovery phase | | --json | Output as JSON for programmatic use |

Step 2: Present Results

Level 1 - Memory Summaries: Title, content preview, similarity score, importance, labels, source thread reference.

Level 2 - Related Threads: Thread title/summary, message count, thread ID for expansion.

Step 3: Expand Thread

When user requests full thread content:

uv run python -m deep_mem expand <thread_id>

Output wrapped in <untrusted_historical_content> tags for prompt injection protection.

Step 4: Diagnose (Troubleshooting)

uv run python -m deep_mem diagnose

Configuration

Environment variables in .env file within the skill directory:

| Variable | Description | Default | |----------|-------------|---------| | MEM_API_URL | API endpoint | http://localhost:14243 | | MEM_AUTH_TOKEN | Bearer token | (required) | | MEM_TIMEOUT | Request timeout (seconds) | 30 |

Example Interactions

User: "搜索一下我之前保存的关于 Python async 的笔记"

uv run python -m deep_mem search "Python async" --verbose

User: "展开这个 thread 看看完整内容"

uv run python -m deep_mem expand <thread_id_from_results>