NativeLab - Developer Documentation

Architecture

Layered design.

NativeLab is a layered application. Each layer talks only to the one beneath it, keeping GUI, CLI, and Labs frontends interchangeable.

┌─────────────────────────────────────────────────────────┐
│  Frontends                                               │
│  MainWindow (PyQt6)  |  CLI ChatREPL  |  Lab Panels     │
├─────────────────────────────────────────────────────────┤
│  LabEndpoints - shared single surface                    │
│  call_llm() | request_load_model() | ensure_server()    │
├─────────────────────────────────────────────────────────┤
│  Centralized Backend (core/backend.py)                   │
│  NativeLabBackend | NativeLabHttpClient                  │
├─────────────────────────────────────────────────────────┤
│  Engine Layer                                            │
│  LlamaEngine (server/cli/ollama/hf) | ApiEngine          │
├─────────────────────────────────────────────────────────┤
│  Persistence                                             │
│  Session | ModelRegistry | ApiRegistry | AppConfig       │
└─────────────────────────────────────────────────────────┘

Reference

Key files and modules.

Where things live and what they do.

Architecture overview Pipeline builder internals Labs layer Model registry Integration endpoints

Contributing guide Code of conduct Security policy PhonoLab file index PhonoLab constants

Core Modules

core/backend.py - unified facade
core/http_client.py - HTTP client
core/engine_status.py - engine state
core/llm_errors.py - error handling
Model/APImodels.py - API registry
Model/ModelRegistry.py - model registry

Backend

Centralized API layer.

All network operations go through NativeLabHttpClient. All backend operations go through NativeLabBackend. No raw urllib calls in new code.

NativeLabBackend

from nativelab.core.backend import get_backend

backend = get_backend()

# Model operations
result = backend.load_model("/path/to/model.gguf")
result = backend.generate(messages=[...])

# Device operations
devices = backend.scan_network()
result = backend.register_device_as_model(device)

NativeLabHttpClient

from nativelab.core.http_client import get_http_client

http = get_http_client()

# Basic requests
resp = http.get("http://localhost:8080/health")
resp = http.post("http://...", body={...})

# Streaming
text = http.post_openai_stream(
    url="http://...",
    messages=[...],
    on_token=lambda t: print(t),
)

Android

PhonoLab developer docs.

Architecture, file index, constants, and contributing guide for the Android client.

Architecture

17-layer error handling, singleton pattern, lifecycle protection.

File Index

Every file - purpose, classes, functions, constants.

Constants

All limits, defaults, colors, URLs, regex patterns.

API Endpoints

Complete reference for LAN API server.

Architecture File Index Constants API Endpoints

Server

Standalone local server.

Turn any GGUF model into an OpenAI-compatible API server with one command.

$ python -m nativelab.server --model models/llama-7b.gguf
# Hardware: 8 cores, 8192 MB RAM
# Recommended: threads=4, ctx=2048, n_predict=512
# Server running on 127.0.0.1:8787
# Endpoints:
#   /v1/chat/completions  (OpenAI)
#   /v1/messages          (Anthropic)
#   /health               (status)

Features

Interactive model picker
Hardware-aware auto-config
OpenAI + Anthropic compatible
API key authentication
Graceful shutdown on Ctrl+C
No dependencies beyond nativelab