Agenticx Memory Architect

AgenticX Memory Architect

Guide for building agents with persistent memory capabilities.

Overview

AgenticX integrates with Mem0 for long-term memory, providing agents with the ability to remember past interactions, learn from experience, and maintain context across sessions.

Installation

bashCopy code
pip install "agenticx[memory]"
# Includes: mem0, chromadb, qdrant-client, redis, milvus

Memory System Components

Basic Memory Usage

Initialize Memory

pythonCopy code
from agenticx.memory import MemoryManager

memory = MemoryManager(
    provider="mem0",
    config={
        "llm": {"provider": "openai", "config": {"model": "gpt-4"}},
        "vector_store": {"provider": "chroma"}
    }
)

Store and Retrieve

pythonCopy code
# Add a memory
memory.add(
    content="User prefers concise reports with bullet points",
    user_id="user-123",
    agent_id="analyst"
)

# Search memories
results = memory.search(
    query="What format does the user prefer?",
    user_id="user-123"
)
for r in results:
    print(f"[{r.score:.2f}] {r.content}")

# Get all memories for a user
all_memories = memory.get_all(user_id="user-123")

Memory-Enhanced Agents

Attach Memory to an Agent

pythonCopy code
from agenticx import Agent, AgentExecutor
from agenticx.memory import MemoryManager
from agenticx.llms import OpenAIProvider

memory = MemoryManager(provider="mem0")
agent = Agent(
    id="assistant",
    name="Personal Assistant",
    role="Assistant with memory",
    goal="Help users while remembering their preferences",
    organization_id="default"
)

executor = AgentExecutor(
    agent=agent,
    llm=OpenAIProvider(model="gpt-4"),
    memory=memory
)

# First interaction — learns preference
result = executor.run(task_1)

# Later interaction — recalls preference
result = executor.run(task_2)  # agent remembers context from task_1

Memory Extraction

AgenticX can automatically extract memorable facts from conversations:

pythonCopy code
from agenticx.core.memory_extraction import MemoryExtractor

extractor = MemoryExtractor(llm=llm)
facts = extractor.extract(conversation_history)
# facts: ["User is a data scientist", "Prefers Python over R", ...]

for fact in facts:
    memory.add(content=fact, user_id="user-123")

Vector Store Backends

pythonCopy code
# Qdrant example
memory = MemoryManager(
    provider="mem0",
    config={
        "vector_store": {
            "provider": "qdrant",
            "config": {"host": "localhost", "port": 6333}
        }
    }
)

Healthcare Example

pythonCopy code
# Medical knowledge memory
memory.add(
    content="Patient has Type 2 diabetes, diagnosed 2023",
    user_id="patient-456",
    metadata={"category": "medical_history"}
)

# Query with context
results = memory.search(
    query="What chronic conditions does the patient have?",
    user_id="patient-456"
)

CLI Memory Operations

bashCopy code
# Run the memory example
python examples/memory_example.py

# Healthcare scenario
python examples/mem0_healthcare_example.py

Best Practices

1. Scope memories — always associate with user_id and/or agent_id
2. Dedup — check for similar memories before adding
3. TTL — set expiration for time-sensitive information
4. Privacy — never store PII without consent; use data isolation
5. Vector store selection — ChromaDB for dev, Qdrant/Milvus for production
6. Memory extraction — automate fact extraction from conversations
7. Test retrieval — verify that stored memories are actually retrievable