AI MODES COMPARED

Which AI mode should you use?

QDev can write jq four different ways. They're not equal. Here's how QDev Pro, WebLLM, Chrome's built-in model, and Ollama stack up — and why Pro is the one we recommend for almost everyone.

TL;DR

Use QDev Pro.

Nothing to install, no GPU, best jq accuracy, and it works identically on Chrome and Firefox. The local modes exist for specific situations — offline, or a hard requirement that nothing at allleaves the device — not because they're better day to day.

At a glance

CRITERION	QDEV PRO ✦ hosted · recommended	WEBLLM local · WebGPU	CHROME BUILT-IN local · Gemini Nano	OLLAMA local · self-run
Setup	✓ Sign in no install, no download	△ Auto-download multi-GB weights on first use	✓ None browser-managed	✗ Install + run local server + model pull
Works on Chrome & Firefox	✓ Both, identical	✗ Chrome only not Firefox on macOS	✗ Chrome only	△ Both needs the local app
Hardware needed	✓ Any device runs server-side	✗ Capable GPU WebGPU required	△ Recent Chrome + Nano available	✗ Strong local machine
jq accuracy	✓ Best tuned for jq	△ Good weaker on hard asks	△ Good small on-device model	△ Good–best depends on your model
Speed & consistency	✓ Fast, consistent	△ Varies by GPU + per-session warm-up	△ Varies	△ Varies by machine
Always the current model	✓ Yes we keep it updated	✗ Manual re-download to update	△ Browser-managed	✗ Manual you pull updates
Data that leaves the device	△ Schema + question never your values	✓ Nothing	✓ Nothing	✓ Nothing
Price	QDev Pro see pricing →	✓ Free	✓ Free	✓ Free
Verdict	✓ Recommended for almost everyone	△ Niche privacy-absolute, Chromium	△ Chrome-only casual use	△ Advanced if you already run it

Every mode sends your values nowhere. QDev Pro sends only your schema + question; local modes send nothing at all.

Why QDev Pro wins for almost everyone

Zero setup, any machine

Sign in and go — no multi-gigabyte model download, no WebGPU or GPU dependency, nothing to install. It runs the same on a maxed-out workstation and a base laptop.

The same on Chrome and Firefox

Local WebGPU models don't run on Firefox for macOS at all. QDev Pro is one path, identical on both browsers, so your team isn't split by hardware.

Best jq accuracy

A larger hosted model we keep tuned for jq handles the tricky reshapes and ambiguous asks the small on-device models miss — and it's fast and consistent.

Always current, nothing to manage

No re-downloads, no version drift, no VRAM warm-up. You get the model we're running today, and it improves without you doing anything.

What the local modes are actually for

They're not the everyday choice — they're escape hatches for specific constraints. Here's where each one earns its place.

WebLLM

PRIVACY-ABSOLUTE · CHROMIUM

Runs a ~3B model in your browser via WebGPU. Genuinely nothing leaves the device — the strongest privacy posture there is.

GOOD FOR

Work under a strict no-data-egress policy — once the model is downloaded, nothing you query ever leaves — on a Chromium browser with a capable GPU.

THE CATCH

Multi-GB first-run download, a per-session GPU warm-up, lower accuracy on hard queries, and it can't run on Firefox for macOS.

Chrome built-in

CHROME ONLY

Uses Chrome's on-device Gemini Nano through the Prompt API. No download, browser-managed, and fully local.

GOOD FOR

Casual, occasional asks for people already on a recent Chrome with the model available.

THE CATCH

Chrome-only, so it doesn't unify your browsers, and the small model trails QDev Pro on anything non-trivial.

Ollama

ADVANCED / POWER USER

Points QDev at a local Ollama server for the strongest local models available.

GOOD FOR

Power users who already run Ollama and want a specific local model with no egress.

THE CATCH

Requires installing and running a local server plus pulling models — the opposite of the no-setup goal.

● THE PRIVACY GAP IS SMALLER THAN IT LOOKS

People reach for local modes to keep data private — but QDev Pro already never sends your values. Only the schema (field names & types) and your question leave, through a stateless, zero-retention proxy. For the vast majority of work that's the same practical privacy as local, without the download, the GPU, or the Firefox gaps. Choose local when policy demands literally-nothing-leaves — not by default.

The bottom line

✓QDev Pro is what we recommend: least friction to set up, strongest results across the board.

→Pick WebLLM only when policy requires that nothing at all leaves the device, and you're on Chromium with a GPU.

→Chrome built-in is fine for light, occasional use if you live in Chrome.

→Ollama makes sense only if you already run it and want a specific local model.

Start with Pro. Switch later if you ever need to.

14-day trial, no card. The local modes are always there — free — if your situation calls for them.

See pricing How it works

QDev

local AI is free · hosted is QDev Pro