LocalPilot Logo LocalPilot v0.1.1
GitHub
⚙️ Deep Dive & Core Principles

Architecture & Safety Design

How LocalPilot cleanly isolates inference backends, external execution services, and deployment methods without compromising user safety.

Clean Abstractions

The 3-Axis Architecture Model

Instead of treating local AI as a flat "runtime" black box, LocalPilot explicitly separates execution into three independent, validated dimensions:

Axis A

Inference Backend

The raw engine or computation library performing tensor arithmetic.

  • • llama.cpp
  • • vLLM (Advanced)
  • • Service-managed
Axis B

Execution Service

The product or daemon that manages model life-cycles, downloads, and endpoints.

  • • Ollama
  • • LM Studio / llmster
  • • Docker Model Runner
Axis C

Deployment Method

How the computation process is hosted and controlled by LocalPilot.

  • • Direct worker process
  • • Local API service
  • • Container
  • • Browser
Why this matters to the user: LocalPilot will never assume an endpoint reveals the backend, never hijack an externally managed daemon, and never modify your system without explicit capability validation.

Frequently Asked Questions

Do I need a gaming PC or an expensive dedicated GPU?

No. Modern 3B models (like Llama 3.2 3B or Phi-3 Mini) run smoothly on ordinary Intel and AMD laptop CPUs. If you own an Apple Silicon Mac (M1/M2/M3/M4), memory is unified between the CPU and GPU, delivering rapid performance without extra hardware.

Will LocalPilot delete or touch my existing Ollama or LM Studio models?

No. LocalPilot enforces strict process and storage ownership boundaries. External services discovered on your computer are treated with respect: LocalPilot can connect and talk to them, but will never stop external daemons, delete external files, or change configuration outside its managed directory.

What is Docker, and do I need to install it?

Docker is a virtualization tool used by developers to package software into isolated containers. You do not need Docker to use LocalPilot. LocalPilot runs directly and natively on your computer. If you happen to have Docker installed, LocalPilot can optionally detect the Docker Model Runner, but it is strictly optional.

Does LocalPilot send any telemetry or conversation history online?

Zero. LocalPilot has no telemetry, analytics, user tracking, or advertising SDKs. All network calls are strictly limited to downloading model artifacts from public registries when you explicitly request them. Local inference runs completely offline.

Where does LocalPilot store downloaded models?

All managed files reside in your operating system's standard user data directory (~/.local/share/localpilot on Linux or ~/Library/Application Support/localpilot on macOS). You can view total space or reclaim it with a single command at any time.