LOCAL_AI_STACK
LOCAL_AI_STACK / start
v2.2 · 2026-06-12
SYNOPSIS
A beginner-safe path for choosing a local AI setup. Avoids "best" claims. Gives conservative steps and links to records that show their verification status.
QUICK_ANSWER
Start with your hardware, not the tool. Local AI depends on RAM, VRAM, Mac unified memory, model size, quantization, context length, and whether your workflow stays local. Once you know your hardware tier, choose one first app, run one small model, and only then move into larger models, PDF chat, Open WebUI, or private workflows.
STEPS
[01]
CHECK_HARDWARE
Start with an estimate before downloading large models.
→/tools/ram-vram-calculator
[02]
GET_STACK_SUGGESTION
Use deterministic guidance before comparing runtime records.
→/tools/stack-recommender
[03]
COMPARE_TOOLS
Filter GUI, API, document, Docker, and privacy caveats across local AI tools.
→/tools/compare
[04]
READ_PRIVACY_GUIDE
Local is not automatically private unless the whole workflow stays local.
→/guides/local-ai-privacy-guide
[05]
INSTALL_ONE_TOOL
Use a source-backed guide and stop when something is unclear.
→/guides/install-lm-studio
RECOMMENDED_FIRST_PATHS

Path A: Easiest desktop app

Use LM Studio if you want a graphical app for downloading models and chatting locally. Read Install LM Studio.

Path B: Best local runtime path

Use Ollama if you want a lightweight runtime, local API, or a base for Open WebUI later. Read Install Ollama.

Path C: Browser-style local workspace

Install Ollama first, then add Open WebUI. Read Open WebUI with Ollama.

Path D: Private document testing

Start with the privacy guide, then read Chat With PDFs Locally. Do not upload confidential documents until you know where the model, embeddings, files, and logs live.

AVOID_THESE_FIRST
  • Do not download the biggest model first.
  • Do not assume “local app” means “local model.”
  • Do not assume 8GB system RAM is the same as 8GB GPU VRAM.
  • Do not expose a local server to your network until you understand the security model.
  • Do not use sensitive documents for the first test run.