t0ken-memoryx 2.1.1

MemoryX Python SDK - Give your AI agents long-term memory

🧠 MemoryX Python SDK

Give your AI agents long-term memory in 2 lines of code.

MemoryX is a cloud-native memory layer for AI agents. It provides semantic search, automatic categorization, and project-level memory sharing across multiple agents.

✨ Features

• 🚀 Zero Config - Auto-registration, no API key management needed
• 🔍 Semantic Search - Find memories by meaning, not keywords
• 🤖 Multi-Agent Support - Multiple agents share memories within a project
• 📊 Auto Categorization - Memories are automatically classified (semantic, episodic, etc.)
• 🔐 Secure by Design - Machine fingerprint authentication, no passwords
• ☁️ Cloud Sync - Memories stored securely in the cloud
• 📦 Zero Dependencies - Uses only Python standard library

📦 Installation

pip install t0ken-memoryx

🚀 Quick Start

from memoryx import connect_memory

# Connect - auto-registers on first use
memory = connect_memory()

# Store memories (async, non-blocking)
memory.send_memories([
    {"content": "User prefers dark mode in VSCode"},
    {"content": "User's project uses TypeScript and React"},
])

# Search memories semantically
results = memory.search("coding preferences")
for m in results["data"]:
    print(f"- {m['content']} (score: {m['score']})")

🎯 Use Cases

AI Assistants

# Remember user preferences across sessions
memory.send_memories([
    {"content": "User prefers concise answers without code comments"},
    {"content": "User works in fintech industry"},
])

# Later, in another session...
results = memory.search("communication style")

Multi-Agent Systems

# Agent 1: VSCode Extension
memory_vscode = APIClient()
memory_vscode.auto_register("vscode", "my-extension")
memory_vscode.send_memories([{"content": "User is debugging auth module"}])

# Agent 2: CLI Tool (same machine, shares project)
memory_cli = APIClient()
memory_cli.auto_register("cli", "my-cli")
# Can search memories stored by VSCode agent!
results = memory_cli.search("debugging")

Conversation Memory

# Store conversation for later analysis
memory.send_conversation("conv-123", [
    {"role": "user", "content": "I need help with my React app"},
    {"role": "assistant", "content": "Sure, what's the issue?"},
    {"role": "user", "content": "The useState hook isn't updating"},
])

📖 API Reference

connect_memory(base_url=None, verbose=True)

Quick connect with auto-registration.

from memoryx import connect_memory

memory = connect_memory()  # First time: auto-registers
# Output: MemoryX activated
#         Agent ID: openclaw-abc12345

APIClient(config)

Advanced usage with explicit configuration.

from memoryx import APIClient

client = APIClient({
    "api_key": "mx_ope_xxx",  # Optional: use existing key
    "api_base_url": "https://t0ken.ai/api"  # Optional: self-hosted
})

# Auto-register new agent
result = client.auto_register("my-agent", "Agent Name")
print(result["agent_id"])  # my-agent-abc12345
print(result["api_key"])   # mx_myagent_xxx

send_memories(memories)

Store memories (batch supported).

# Single memory
result = client.send_memories([
    {"content": "User likes pizza", "metadata": {"category": "preference"}}
])

# Batch memories
result = client.send_memories([
    {"content": "User's timezone is UTC+8"},
    {"content": "User speaks English and Chinese"},
])
print(result["task_id"])  # Async task ID

search(query, limit=10)

Semantic search across all project memories.

results = client.search("food preferences", limit=5)
for m in results["data"]:
    print(f"- {m['content']} (category: {m['category']}, score: {m['score']})")

list(limit=50, offset=0)

List memories with pagination.

memories = client.list(limit=20, offset=0)
print(f"Total: {memories['total']}")

delete(memory_id)

Delete a memory.

client.delete("memory_abc123")

get_task_status(task_id)

Check async task status.

status = client.get_task_status("task_abc123")
print(status["status"])  # PENDING, STARTED, SUCCESS, FAILURE

get_quota()

Get account quota.

quota = client.get_quota()
print(f"Plan: {quota['quota']['tier']}")
print(f"Searches remaining: {quota['quota']['cloud_search']['remaining']}")

🔒 Security

• Machine Fingerprint: Uses MAC address + hardware info for authentication
• No Passwords: API keys are auto-generated and tied to agents
• Project Isolation: Each machine gets its own project, memories are isolated

🌐 Self-Hosted

from memoryx import connect_memory

memory = connect_memory(base_url="http://your-server:8000/api")

📚 More Resources

• Documentation
• Node.js SDK
• GitHub

📄 License

MIT © MemoryX Team