cognimemo-memory（认知记忆）

CogniMemo - Universal AI Memory

CogniMemo provides persistent, intelligent memory for AI applications. Unlike session-based memory that disappears, CogniMemo stores, understands, and learns from interactions over time.

Why CogniMemo?

• Cross-app memory - Same memory across ChatGPT, Claude, Gemini, DeepSeek
• Model-agnostic - Works with OpenAI, Anthropic, Gemini, Mistral, Ollama
• Auto-captured - Decides what matters, no manual organization
• Permission-based - Users control what each app can access
• Simple API - REST API, SDKs, LangChain adapters

How It Works

1. Memory Auto-Captured

CogniMemo captures from:

• Chat conversations
• Documents and links
• Tasks, decisions, notes
• User actions

2. AI Understands Context

Extracts:

• Entities (people, places, things)
• Relationships
• Patterns and habits
• Temporal context

3. Permission-Based Access

• Apps see only approved memory types
• Users can revoke access anytime
• Scoped by permission level

Quick Start

Step 1: Get API Key

1. Go to https://cognimemo.com
2. Create account
3. Generate API key from dashboard
4. Add to environment:

COGNIMEMO_API_KEY=your-api-key-here

Step 2: Install SDK

# Python
pip install cognimemo

# Node.js
npm install @cognimemo/sdk

Step 3: Initialize Client

from cognimemo import CogniMemo

# Initialize with API key
memory = CogniMemo(api_key="your-api-key")

# Or from environment
memory = CogniMemo()  # Uses COGNIMEMO_API_KEY

Core Operations

Store Memory

# Store a conversation
memory.store(
    user_id="user-123",
    content="User prefers Portuguese language responses",
    metadata={
        "type": "preference",
        "source": "chat",
        "confidence": 0.9
    }
)

# Store a decision
memory.store(
    user_id="user-123",
    content="Decided to use React for the frontend project",
    metadata={
        "type": "decision",
        "project": "web-app",
        "timestamp": "2026-03-16"
    }
)

# Store a task
memory.store(
    user_id="user-123",
    content="Need to prepare quarterly report by Friday",
    metadata={
        "type": "task",
        "deadline": "2026-03-20",
        "priority": "high"
    }
)

Retrieve Memory

# Semantic search
results = memory.search(
    user_id="user-123",
    query="What are the user's preferences?",
    limit=10
)

# Get specific type
preferences = memory.get_by_type(
    user_id="user-123",
    memory_type="preference"
)

# Get recent
recent = memory.get_recent(
    user_id="user-123",
    hours=24
)

Update Memory

# Update existing memory
memory.update(
    memory_id="mem-456",
    content="User prefers concise Portuguese responses",
    metadata={"confidence": 1.0}
)

# Add context to existing memory
memory.append(
    memory_id="mem-456",
    additional_context="Also prefers bullet points over paragraphs"
)

Delete Memory

# Delete specific memory
memory.delete(memory_id="mem-456")

# Clear all memories for a user
memory.clear(user_id="user-123")

# Clear by type
memory.clear(user_id="user-123", memory_type="task")

Memory Types

Permission Scopes

# Request specific permissions
auth_url = memory.get_auth_url(
    scopes=["preferences", "decisions", "tasks"],
    redirect_uri="https://your-app.com/callback"
)

# Check user permissions
permissions = memory.get_permissions(user_id="user-123")
# Returns: {"preferences": True, "decisions": True, "tasks": False}

Integration with AI Models

OpenAI / ChatGPT

import openai
from cognimemo import CogniMemo

memory = CogniMemo()
user_id = "user-123"

# Get relevant context
context = memory.search(
    user_id=user_id,
    query="User preferences and recent decisions",
    limit=5
)

# Build prompt with memory
messages = [
    {"role": "system", "content": f"Context: {context}"},
    {"role": "user", "content": "Help me with my project"}
]

response = openai.chat.completions.create(
    model="gpt-4",
    messages=messages
)

# Store important info from conversation
memory.store(
    user_id=user_id,
    content="User asked about React component library",
    metadata={"type": "context", "session": "current"}
)

Anthropic / Claude

import anthropic
from cognimemo import CogniMemo

memory = CogniMemo()
user_id = "user-123"

# Get memory context
context = memory.search(
    user_id=user_id,
    query="User preferences",
    limit=10
)

client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    system=f"Remember: {context}",
    messages=[{"role": "user", "content": "What should I work on?"}]
)

LangChain Integration

from langchain.memory import CogniMemoMemory
from langchain.chains import ConversationChain
from langchain.llms import OpenAI

# Use CogniMemo as LangChain memory
memory = CogniMemoMemory(
    api_key="your-api-key",
    user_id="user-123"
)

chain = ConversationChain(
    llm=OpenAI(),
    memory=memory
)

# Memory automatically stored and retrieved
response = chain.predict(input="What did we discuss last time?")

OpenClaw Integration

# In OpenClaw skill or agent
from cognimemo import CogniMemo

class CogniMemoTool:
    """Tool for OpenClaw agents to access persistent memory."""
    
    def __init__(self, user_id: str):
        self.memory = CogniMemo()
        self.user_id = user_id
    
    def remember(self, content: str, memory_type: str = "context"):
        """Store something in memory."""
        self.memory.store(
            user_id=self.user_id,
            content=content,
            metadata={"type": memory_type}
        )
        return f"Remembered: {content}"
    
    def recall(self, query: str):
        """Search memory for relevant information."""
        results = self.memory.search(
            user_id=self.user_id,
            query=query,
            limit=10
        )
        return results
    
    def get_preferences(self):
        """Get user preferences."""
        return self.memory.get_by_type(
            user_id=self.user_id,
            memory_type="preference"
        )

Storage Backends

CogniMemo supports multiple storage layers:

Configure via environment:

COGNIMEMO_STORAGE=pinecone  # or weaviate, postgres, redis
COGNIMEMO_PINECONE_API_KEY=your-key
COGNIMEMO_PINECONE_ENV=us-west1-gcp

Best Practices

1. Store Wisely

# Good: Specific, structured memory
memory.store(
    user_id="user-123",
    content="User prefers dark mode in code editors",
    metadata={"type": "preference", "category": "ui"}
)

# Bad: Vague, unstructured
memory.store(user_id="user-123", content="user likes stuff")

2. Search Effectively

# Use semantic queries
results = memory.search(
    user_id="user-123",
    query="What editor preferences does the user have?",
    limit=5
)

3. Respect Privacy

# Check permissions before storing
if memory.has_permission(user_id, "preferences"):
    memory.store(...)

Pricing

• Free Tier: 1,000 memories/month
• Pro: $29/month for 50,000 memories
• Enterprise: Custom pricing for unlimited

Resources

• Website: https://cognimemo.com
• Documentation: https://docs.cognimemo.com
• API Reference: https://api.cognimemo.com/docs
• GitHub: https://github.com/cognimemo/sdk

Error Handling

from cognimemo import CogniMemo, CogniMemoError

try:
    memory.store(user_id="user-123", content="Important info")
except CogniMemoError as e:
    if e.code == "quota_exceeded":
        print("Free tier limit reached. Upgrade at cognimemo.com/pricing")
    elif e.code == "permission_denied":
        print("User has not granted permission for this memory type")
    else:
        raise