Memori Extension Skill

Memory augmentation and LLM call interception using the Memori Python library with optional Zhipu AI API integration.

Overview

This skill provides memory augmentation and LLM call interception capabilities using the Memori Python library. It enables agents to retrieve relevant knowledge from a memory database and inject it into conversations.

Security & Privacy Notice

⚠️ Important: This skill performs the following operations:

File Operations (Always Enabled)

- Read/Write: Local SQLite database (default: ./memori.db)
Read/Write: Optional tech terms configuration file (default: ./config/tech_terms.txt)
These operations are entirely local and do not transmit data externally.

External API Calls (Optional - Requires Explicit Configuration)

- Condition: Only if ZHIPUAI_API_KEY environment variable is set
Data Transmitted: Conversation text (user messages, system prompts, and assistant responses) is sent to Zhipu AI's servers for analysis and augmentation
Purpose: To enhance conversation understanding using Zhipu AI's language models
Control: If ZHIPUAI_API_KEY is NOT set, the skill operates 100% locally with NO external network calls

Recommendations

1. Start Local-Only: Test the skill without ZHIPUAI_API_KEY first to verify local functionality
Explicit Consent: Only set ZHIPUAI_API_KEY if you explicitly consent to sending conversation data to external services
Review Data: Consider what conversation data you're comfortable transmitting before enabling external API features
Sandbox Testing: If providing API keys, test in a sandbox environment first to understand the data flow

Dependencies

This skill requires the following Python packages:

CODEBLOCK0

The memori library is licensed under Apache License Version 2.0.

Optional Dependencies

For Zhipu API augmentation (optional):

CODEBLOCK1

Warning: Installing zhipuai and providing ZHIPUAI_API_KEY enables conversation content to be sent to external servers.

Environment Variables

All Supported Environment Variables

Variable	Description	Required	Default	Privacy Note
INLINECODE8	Zhipu AI API key for conversation augmentation	No	-	⚠️ Enables external API calls - conversation text will be sent to Zhipu AI servers
INLINECODE9

Privacy & Data Flow Notes:

- ⚠️ External API: Only ZHIPUAI_API_KEY enables external network calls. All other variables control local file operations.
✅ Local-Only Mode: If you omit ZHIPUAI_API_KEY, the skill operates 100% locally with no external data transmission.
📁 File Operations: The skill reads/writes the database (memori.db) and optionally the tech terms file (tech_terms.txt). These are stored on your local filesystem.
🔒 Best Practice: Start without ZHIPUAI_API_KEY to test local functionality. Only enable external API if you need enhanced features and consent to the data transmission.

Configuration Examples

System environment:
CODEBLOCK2

OpenClaw configuration (openclaw.json):
CODEBLOCK3

Technical terms file (optional):
CODEBLOCK4

Quick Start

Method 1: Direct Memori Library (Recommended)

CODEBLOCK5

Method 2: Skill Convenience API

CODEBLOCK6

API Reference

Memori Class

`init(db_path, entity_id)`

Initialize Memori instance.

Parameters:

- db_path (str | Path, optional): Database path
INLINECODE25 (str, optional): Entity ID, default INLINECODE26

`search(query, limit, entity_id)`

Search for relevant memories.

Returns: INLINECODE28

`augment(query, limit, entity_id)`

Augment query with retrieved memories.

Returns: INLINECODE30

`store(content, entity_id, metadata)`

Store a new memory.

Returns: int - Memory ID

`get_stats(entity_id)`

Get statistics.

Returns: INLINECODE34

`close()`

Close database connection.

Memory Class

Memory object with attributes:

- id (int): Memory ID
INLINECODE37 (str): Entity ID
INLINECODE38 (str): Memory content
INLINECODE39 (str): Creation timestamp
INLINECODE40 (dict, optional): Metadata

AugmentedContext Class

Augmented context with:

- original_query (str): Original query
INLINECODE42 (List[Memory]): Retrieved memories
INLINECODE43 (str): Augmented prompt
INLINECODE44 (bool): Whether memories were retrieved
INLINECODE45 (int): Number of retrieved memories

Configuration

Database

Default database path: INLINECODE46

The skill will:

- ✅ Read from the database to retrieve memories
✅ Write to the database when storing new memories
✅ Create the database file if it doesn't exist

Technical Terms

The skill uses configurable technical terms for LLM call interception. You can customize these via:

1. Environment variable (comma-separated):
CODEBLOCK7

2. Configuration file (one term per line):
CODEBLOCK8

Note: If neither is set, the skill will still work but may not intercept technical queries as effectively.

Security Considerations

File Operations

This skill performs the following file operations:

Operation	File	Description
Read	INLINECODE47 (default)	Retrieve stored memories
Write

External API Calls

⚠️ Important: If ZHIPUAI_API_KEY is set, this skill may send conversation text to Zhipu AI's servers for augmentation.

To disable external API calls:

- Simply don't set INLINECODE52
The skill will work normally using local memory retrieval only

Recommendations

1. Review the code before enabling external API features
Test in a sandbox first if providing API keys
Use file permissions to protect database and config files
Only enable features you explicitly need

File Structure

CODEBLOCK9

License

This skill is licensed under Apache License Version 2.0.

This skill uses the memori Python library, which is also licensed under Apache License Version 2.0.

Attribution

This skill incorporates the Memori Python library, which is licensed under the Apache License 2.0.
See the LICENSE file for the complete license text.

Memori Library:

- Copyright 2025 Memori Team
License: Apache License 2.0
Repository:

Memori 扩展技能

使用 Memori Python 库实现记忆增强和 LLM 调用拦截，可选配智谱 AI API 集成。

概述

本技能利用 Memori Python 库提供记忆增强和 LLM 调用拦截功能。它使智能体能够从记忆数据库中检索相关知识并将其注入对话中。

安全与隐私声明

⚠️ 重要提示：本技能执行以下操作：

文件操作（始终启用）

- 读取/写入：本地 SQLite 数据库（默认路径：./memori.db）
读取/写入：可选的技术术语配置文件（默认路径：./config/tech_terms.txt）
这些操作完全在本地进行，不会向外部传输数据。

外部 API 调用（可选 - 需要显式配置）

- 条件：仅当设置了 ZHIPUAIAPIKEY 环境变量时
传输数据：对话文本（用户消息、系统提示和助手回复）将被发送至智谱 AI 服务器进行分析和增强
目的：利用智谱 AI 的语言模型增强对话理解能力
控制：如果未设置 ZHIPUAIAPIKEY，本技能将 100% 在本地运行，不进行任何外部网络调用

建议

1. 先本地运行：在不设置 ZHIPUAIAPIKEY 的情况下测试技能，验证本地功能
明确同意：仅在您明确同意将对话数据发送至外部服务时才设置 ZHIPUAIAPIKEY
审查数据：在启用外部 API 功能前，考虑您愿意传输哪些对话数据
沙箱测试：如果提供 API 密钥，先在沙箱环境中测试以了解数据流向

依赖项

本技能需要以下 Python 包：

bash
pip install memori

memori 库采用 Apache License Version 2.0 许可。

可选依赖项

用于智谱 API 增强（可选）：

bash
pip install zhipuai

警告：安装 zhipuai 并提供 ZHIPUAIAPIKEY 将使对话内容被发送至外部服务器。

环境变量

所有支持的环境变量

变量	描述	必需	默认值	隐私说明
ZHIPUAIAPIKEY	用于对话增强的智谱 AI API 密钥	否	-	⚠️ 启用外部 API 调用 - 对话文本将被发送至智谱 AI 服务器
ZHIPUAIMODEL

隐私与数据流说明：

- ⚠️ 外部 API：仅 ZHIPUAIAPIKEY 启用外部网络调用。所有其他变量控制本地文件操作。
✅ 仅本地模式：如果省略 ZHIPUAIAPIKEY，本技能将 100% 在本地运行，不进行外部数据传输。
📁 文件操作：本技能读取/写入数据库（memori.db）和可选的技术术语文件（techterms.txt）。这些文件存储在您的本地文件系统中。
🔒 最佳实践：先在不设置 ZHIPUAIAPI_KEY 的情况下测试本地功能。仅在需要增强功能且同意数据传输时才启用外部 API。

配置示例

系统环境：
bash

可选：启用智谱 API 增强

export ZHIPUAIAPIKEY=your-api-key
export ZHIPUAI_MODEL=glm-4.7

可选：自定义技术术语

export MEMORITECHTERMS=FFI,Rust,Linux,kernel,spinlock

可选：使用自定义数据库路径

export MEMORIDBPATH=/path/to/memori.db

OpenClaw 配置（openclaw.json）：
json
{
skills: {
entries: {
memori-extension: {
enabled: true,
env: {
ZHIPUAIAPIKEY: your-api-key,
ZHIPUAI_MODEL: glm-4.7,
MEMORITECHTERMS: FFI,Rust,Linux,kernel,
MEMORIDBPATH: ./memori.db
}
}
}
}
}

技术术语文件（可选）：
bash

创建配置目录

mkdir -p config

创建术语文件

cat > config/tech_terms.txt << EOF FFI Rust Linux kernel spinlock mutex unsafe EOF

设置环境变量

export MEMORITECHTERMSFILE=config/techterms.txt

快速开始

方法 1：直接使用 Memori 库（推荐）

python
from memori import Memori

memori = Memori(
db_path=memori.db,
entity_id=knowledge-base
)

搜索记忆

memories = memori.search(query, limit=5)

增强查询

context = memori.augment(如何处理自旋锁冲突？, limit=3)

存储记忆

memory_id = memori.store(新内容)

获取统计信息

stats = memori.get_stats()

关闭连接

memori.close()

方法 2：技能便捷 API

python
from skills.memoriextension import search, augment, interceptllm

搜索

memories = search(FFI 绑定, limit=5)

增强

enhanced = augment(如何处理自旋锁冲突？) if enhanced: print(enhanced)

拦截 LLM

messages = [{role: user, content: FFI 问题}] enhanced = intercept_llm(messages)

API 参考

Memori 类

init(dbpath, entityid)

初始化 Memori 实例。

参数：

- dbpath（str | Path，可选）：数据库路径
entityid（str，可选）：实体 ID，默认值为 default

search(query, limit, entity_id)

搜索相关记忆。

返回： List[Memory]

augment(query, limit, entity_id)

使用检索到的记忆增强查询。

返回： AugmentedContext

store(content, entity_id, metadata)

存储新记忆。

返回： int - 记忆 ID

getstats(entityid)

获取统计信息。

返回： dict

close()

关闭数据库连接。

Memory 类

具有以下属性的记忆对象：

- id（int）：记忆 ID
entityid（str）：实体 ID
content（str）：记忆内容
createdat（str）：创建时间戳
metadata（dict，可选）：元数据

AugmentedContext 类

增强上下文包含：

- originalquery（str）：原始查询
retrievedmemories（List[Memory]）：检索到的记忆
enhancedprompt（str）：增强后的提示
hasmemories（bool）：是否检索到记忆
memories_count（int）：检索到的记忆数量

配置

数据库

默认数据库路径：./memori.db

本技能将：

- ✅ 读取数据库以检索记忆
✅ 写入数据库以存储新记忆
✅ 创建数据库文件（如果不存在）

技术术语

本技能使用可配置的技术术语进行 LLM 调用拦截。您可以通过以下方式自定义：

1. 环境变量（逗号分隔）：
bash
export MEMORITECHTERMS=FFI,Rust,Linux,kernel,spinlock,mutex,unsafe

2. 配置文件（每行一个术语）：
bash

创建配置文件

mkdir -p config
cat > config/tech_terms.txt << EOF
FFI
Rust
Linux
kernel
spinlock
mutex
unsafe
EOF

设置环境变量

export MEMORITECHTERMSFILE=config/techterms.txt

注意：如果两者均未设置，技能仍可

memori-extension记忆扩展