## Prerequisites

- Python 3.8+ or Node.js 18+
- A Mengram API key — [get one here](/#signup)
- Any LLM API (OpenAI, Anthropic, etc.) or a local model

## Step 1: Install

### Python

```
pip install mengram-ai
```

### JavaScript

```
npm install mengram
```

## Step 2: Initialize

### Python

```
from mengram import Mengram

m = Mengram(api_key="mg-...")  # or set MENGRAM_API_KEY env var
```

### JavaScript

```
import Mengram from 'mengram';

const m = new Mengram({{ apiKey: 'mg-...' }});
```

## Step 3: Store memories after each conversation

After your agent finishes a conversation turn, pass the exchange to Mengram. It automatically extracts [all three memory types](/content/blog/semantic-episodic-procedural-memory/index.html).

### Python

```
# Store the conversation — Mengram extracts facts, events, and workflows
m.add(
    "User asked how to deploy to production. I walked them through "
    "the CI/CD pipeline: push to main, GitHub Actions runs tests, "
    "builds Docker image, deploys to staging, then promotes to prod.",
    user_id="user-123"
)
```

### JavaScript

```
await m.add(
  "User asked how to deploy to production. I walked them through " +
  "the CI/CD pipeline: push to main, GitHub Actions runs tests, " +
  "builds Docker image, deploys to staging, then promotes to prod.",
  {{ userId: 'user-123' }}
);
```

## Step 4: Search memories before responding

```
# Python
results = m.search("deployment process", user_id="user-123")
for r in results:
    print(r.memory, r.type, r.score)
```

```javascript
// JavaScript
const results = await m.search('deployment process', {{ userId: 'user-123' }});
results.forEach(r => console.log(r.memory, r.type, r.score));
```

## Step 5: Use Cognitive Profile for instant personalization

Instead of searching for specific memories, generate a complete system prompt:

```
# Python — one API call returns a ready-to-use system prompt
profile = m.profile(user_id="user-123")
print(profile)
# "You are assisting user-123, a developer who works with CI/CD pipelines..."

# Use it with any LLM
response = openai.chat.completions.create(
    model="gpt-4o",
    messages=[\
        {{"role": "system", "content": profile}},\
        {{"role": "user", "content": user_message}}\
    ]
)
```

## Full example: OpenAI agent with memory

```
from openai import OpenAI
from mengram import Mengram

openai = OpenAI()
m = Mengram()

def chat(user_id: str, message: str) -> str:
    # Get personalized system prompt from memory
    profile = m.profile(user_id=user_id)

response = openai.chat.completions.create(
        model="gpt-4o",
        messages=[\
            {{"role": "system", "content": profile}},\
            {{"role": "user", "content": message}}\
        ]
    )
    reply = response.choices[0].message.content

# Store the exchange in memory
    m.add(f"User: {{message}}\nAssistant: {{reply}}", user_id=user_id)
    return reply
```

That's it. Your agent now remembers every conversation and gets smarter over time. Also works with [CrewAI and LangChain](/content/blog/ai-memory-for-crewai-langchain/index.html), or as an [MCP server for Claude Desktop](/content/blog/mcp-memory-server-setup/index.html).