latzero 0.5.0

Zero-latency, zero-fuss shared memory for Python — dynamic, encrypted, and insanely fast.

latzero

Zero-latency, zero-fuss shared memory for Python — dynamic, encrypted, and insanely fast.

🚀 170,000+ ops/sec | ⚡ 0.006ms latency | 🔒 AES-256 encryption

Overview

latzero is a Python package designed to make inter-process communication (IPC) and shared-memory data exchange effortless. Unlike traditional shared memory systems that require fixed buffer sizes and manual serialization, latzero enables developers to:

• Create dynamic shared-memory pools accessible by multiple processes or clients
• Pass any pickleable object directly — no manual encoding/decoding
• Enable optional encryption and authentication for secure multi-process collaboration

latzero is ideal for AI workloads, distributed systems, and low-latency microservices that need real-time shared state management.

Core Features

Dynamic Shared Memory Pools
No predefined memory size. Pools expand and contract dynamically as new data arrives.

Multi-Client Access
Multiple processes/clients can connect to the same pool simultaneously and share data in real time.

Auto Cleanup
Data can have optional timeouts (auto_clean=5), automatically clearing entries after specified seconds.

Encryption & Authentication
Pools can be protected with passwords. If encryption=True, the password becomes the encryption key.

Data-Type Preservation
Stored data retains its type (int, str, dict, etc.) across clients.

High-Performance Batched Writes
Use set_fast() + flush() for 77,000+ writes/sec. Standard set() syncs immediately.

Self-Destructing Pools
Pools live only as long as one or more connected processes are active. When all disconnect, the pool is automatically destroyed.

Pickle + msgpack Serialization
Any pickleable Python object can be stored. Use msgpack for 3-5x faster serialization.

Event Hooks
Hooks for client events: on_connect, on_disconnect, on_update, on_delete.

Socket-like Events API
RPC-style emit() and call() for inter-process communication with OS-native signaling (~50μs latency).

Async Support
Full async/await API with AsyncSharedMemoryPool.

CLI Tool
Manage pools from command line: latzero list, latzero inspect, etc.

Installation

pip install latzero

Server Mode

LatZero can still run in standalone shared-memory mode, but it can now also connect to a local TCP server.

Start the server separately:

latzero-server

Then connect from Python:

from latzero import LatZero

client = LatZero("latzero://client-1", pool="alpha")
client.set("user", {"name": "Alice"}, persistent=True)
print(client.get("user"))

Server mode uses JSON-over-TCP on 127.0.0.1:14130 by default, supports named open or auth-guarded pools, explicit buffer subscriptions, targeted app calls, and optional persistent buffers. Server mode values must be JSON-serializable.

Quick Start

Creating a Pool

from latzero import SharedMemoryPool

pool_manager = SharedMemoryPool()
pool_manager.create(
    name="myPool",
    auth=True,
    auth_key="super_secret",
    encryption=True
)

Connecting to a Pool

ipc = pool_manager.connect(
    name="myPool",
    auth_key="super_secret"
)

Basic Operations

# Set values with optional auto-cleanup
ipc.set("key", value, auto_clean=5)

# Retrieve values
result = ipc.get("key")

Type-Safe Example

ipc.set("number", 42)
ipc.set("text", "yo bro")
ipc.set("data", {"a": 1, "b": 2})

print(ipc.get("number"))  # 42 (int)
print(ipc.get("text"))    # "yo bro" (str)
print(ipc.get("data"))    # {"a": 1, "b": 2} (dict)

High-Performance Batched Writes

with pool_manager.connect("myPool", auth_key="super_secret") as client:
    # Use set_fast() for batched writes (100x faster)
    for i in range(1000):
        client.set_fast(f"key_{i}", {"value": i})

    # Flush to persist all writes
    client.flush()

    # Reads are always instant
    print(client.get("key_0"))  # {'value': 0}

Events API (Socket-like IPC)

latzero provides a socket.io-style events API for RPC and fire-and-forget messaging:

from latzero import SharedMemoryPool

pm = SharedMemoryPool()
pm.create("my_pool")

with pm.connect("my_pool") as ipc:
    # Register event handler
    @ipc.on_event("compute:multiply")
    def multiply(x: int, y: int) -> int:
        return x * y

    # Start background listener
    ipc.listen()

    # Fire-and-forget emit
    ipc.emit_event("user:login", username="alice")

    # RPC call with response (~50μs latency)
    result = ipc.call_event("compute:multiply", x=7, y=6, _timeout=1.0)
    print(result)  # 42

    # Namespaced emitters
    compute = ipc.event_emitter("compute")

    @compute.on("add")
    def add(a: int, b: int) -> int:
        return a + b

    compute.listen()
    result = compute.call("add", a=10, b=5)  # 15

System Architecture

Core Components

Memory Controller
Manages shared memory segments dynamically.

Pool Registry
Tracks all active pools via metadata.

Encryption Layer
AES-GCM encryption for secure reads/writes.

Data Layer (Pickle Serializer)
Automatic serialization with zlib compression.

IPC Protocol
Uses multiprocessing.shared_memory for communication.

Auto-Reclaim Daemon
Monitors idle pools and clears them when unused.

Security Model

Concern	Mechanism
Unauthorized access	Password-based authentication
Data leakage	AES-256 encryption when encryption=True
Data tampering	Integrity checked using HMAC
Memory persistence	Pools are ephemeral; memory is released after last client disconnects

Performance

Operation	Throughput	Latency
set_fast() (batched)	77,000+ ops/sec	~0.01ms
get()	287,000+ ops/sec	~0.003ms
set() (immediate sync)	600 ops/sec	~1.6ms
Mixed workload	146,000+ ops/sec	~0.007ms

vs Other Methods

Method	Throughput	Speedup
latzero	170,000 ops/sec	—
Raw Socket	5,277 ops/sec	32x faster
HTTP (Flask)	133 ops/sec	1,280x faster

Examples

Check the examples/ directory for usage demos:

• simple_pool.py - Basic pool operations
• encrypted_pool.py - Secure pool with encryption
• multi_client_demo.py - Concurrent multi-client access
• events_demo.py - Socket-like events API demo

Dependencies

• multiprocessing.shared_memory
• cryptography (for AES)
• pickle, zlib
• threading, multiprocessing
• psutil (for process detection)

Roadmap

Phase 1: Core shared memory pools + pickle serialization
Phase 2: Auth + encryption
Phase 3: Dynamic memory expansion + auto-clean
Phase 4: Performance optimization + PyPI release
Phase 5: Real-time event hooks, WebSocket bridges

Use Cases

• AI agents sharing memory
• Game servers syncing states
• Local caching for microservices
• High-speed analytics pipelines
• Multi-agent orchestration systems

About

latzero makes shared-memory IPC as easy as Redis, without the network overhead. Fast, simple, encrypted, ephemeral — a zero-latency memory layer for Python developers.

Created by BRAHMAI
https://brahmai.in