# Secure Context Memory with enVector MCP
This document presents an use case for building a **Secure Context Memory for LLMs using enVector MCP**.
## Context Memory for Personalized AI Assistants
Today, many users regularly use conversational AI services such as ChatGPT and other chat-based assistants in their daily lives. While these systems are trained on massive datasets and can generate high-quality responses, they share an inherent limitation: **they cannot retain information beyond their training data**.
To compensate for this limitation, most AI services automatically store user conversation histories and build internal knowledge databases on top of them. By leveraging past conversations, preferences, and contextual signals as **memory**, these systems can deliver more **accurate and personalized** responses over time.
However, this approach introduces a fundamental problem.
User conversations often contain **personal or confidential data**, such as health information, financial records, lifestyle patterns, workplace documents, and sensitive perspectives. Storing such data on servers operated by service providers—such as OpenAI or other AI platforms—creates serious privacy and security risks.
Through numerous real-world security incidents and data breaches, we have learned how dangerous it can be to store personal data on external systems. As a result, there is a growing demand for AI services that deliver personalized experiences **while keeping the data secure**.
## Context Memory with enVector MCP
enVector MCP was designed in response to this challenge.
By using enVector, developers can build personalized AI assistants while ensuring that **sensitive information is used only in encrypted states**.
The core principles of enVector are as follows:
* **Similarity search is performed on encrypted vectors**
* The server performs search operations without knowing the contents of the data
* The results can be decrypted only by the user; the server sees ciphertext only and has no access to plaintext
In other words, the retrieval pipeline used in traditional LLM context memory systems (incl. RAG) can be implemented **entirely over encrypted data**.
enVector MCP acts as the **control plane** governing this secure retrieval process.
This secure context memory architecture is especially valuable in security-sensitive domains such as:
* Personal health and medical records
* Long-term conversational history and mental health context
* Personal asset management and investment data
* Confidential internal enterprise information
## Preparing enVector MCP for LLMs
1. First, to connect enVector MCP to Claude (or your LLM of choice), see [MCP Guide](/integrations/mcp.md).
2. Set Rule Prompts appropriate for your use case and your LLMs, for example,
```
# Context Memory System
You have access to enVector MCP for persistent context memory storage.
Use it to remember important information across conversations.
## How to store and retrieve with enVector MCP
- Use 'my_context_memory' index in enVector MCP. If not exists, use 'create_index' tool with dimension 384.
- To store memory, use 'insert_document_from_text' tool in enVector MCP.
- To retrieve memory, use 'search' tool in enVector MCP.
## When to Store Memory
Store information when:
- User shares personal preferences, goals, or background
- Important decisions or conclusions are made
- Key facts about ongoing projects are discussed
- User explicitly asks you to remember something
...
```
## Context Memory Example
Users naturally share important personal information during everyday conversations.
For example, consider a scenario where a user stores their health checkup results as long-term context memory.
The following example of personal medical record is stored via enVector MCP **in fully encrypted form**:
```
Basic Information
Age: 49
Gender: Male
Occupation: Office worker
Smoking: 10–15 cigarettes per day (20+ years)
Alcohol: 3–4 times per week (1 bottle of soju or 3 cans of beer)
Exercise: Light walking 1–2 times per week
Medical history: Early-stage hypertension
2023 Health Checkup Results
Blood pressure: 142/92 mmHg (mild hypertension)
Weight: 76.5 kg / Height: 172 cm
BMI: 25.9 (borderline overweight)
Total cholesterol: 210 mg/dL
...
Mental health screening: Depression diagnosis
Physician’s notes:
Blood pressure management recommended.
Lifestyle changes (exercise and diet) advised.
Smoking and alcohol reduction recommended.
Continued mental health counseling suggested.
```
This data belongs entirely to the user.
Without enVector, such information would typically be stored directly by the AI service provider. With enVector, however, the server cannot interpret the contents—only encrypted vectors and ciphertext exist.
Later, the user may ask questions such as:
1. “Recommend supplements that would be suitable for me”
2. “Recommend insurance products”
enVector MCP embeds the user’s query and performs **similarity search over the encrypted memory store**. Throughout this process, the server remains unaware of the data’s meaning, and the retrieved results are decrypted only on the user side.
As a result, the AI assistant can generate responses that accurately reflect the user’s health history and personal context—**while ensuring that sensitive personal data is never exposed to external systems**.
no context memory
context memory with enVector
---
# Agent Instructions: Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://docs.envector.io/tutorials/envector-mcp-context-memory.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.