LibShortcut

Vector AI Agent

by Mincofficial v9.3
iOS 17+
Requires
AI
Category
Jun 2026
Updated

Vector is an AI that does, not just chats. The whole agent is built inside Shortcuts. There is no companion app to install from the App Store. It supports more than fifteen models, calls tools on your behalf, and stores every byte of config and history locally on your iPhone in the Files app.

What it actually does

Vector behaves like a small assistant that can take action between replies. You ask for something, the model picks one or more tools, runs them, and folds the results back into the conversation. Tools include sending an iMessage or WhatsApp message, drafting emails, placing calls, searching contacts, setting timers, alarms, and reminders, creating Calendar events and Apple Notes, searching the web and reading pages, running sandboxed Python (64MB of memory, up to 10,000 lines), checking weather, reading the screen with permission, reading the clipboard, generating images, and opening URLs or other shortcuts.

What you’ll need

Adding it to your iPhone

  1. Tap Add Shortcut on this page to open Vector in the Shortcuts app.
  2. Add Vector Config from the same author.
  3. Open Vector once so it can create its folders in Files.
  4. Pick a name, choose text or voice mode, and select a default model during the first-run setup.

First-run setup

The first run walks through four short questions. First it asks for a name Vector should answer to. Next you choose text mode or voice mode as the default interaction style. The third question sets a default model, with Minimax M2.7 recommended as a balanced starting point. Last comes the terms of service. After that, anything you want to change later lives in Vector Config — model, personality, memory, and mode toggles are all reconfigured from there.

Using it day-to-day

Talk or type. Vector reads the request, decides which tools fit, and chains several in one loop when a task needs it. Ask it to text a friend, set a 20-minute timer, then draft a note, and it will work through all three before replying. A notification fires when something runs longer than expected. Voice mode is handy in the kitchen while your hands are wet. The Python tool is good for quick calculations or one-off scripts on the road. Screen reading helps when you want context from whatever app is in front of you.

Picking a model

There are three tiers in Vector Config. Smart includes Minimax M3, Deepseek V4, and Kimi K2.6, suited to harder coding and reasoning problems. Balanced covers Minimax M2.7 and Qwen3.6 Plus, the sensible daily default. Fast covers Step 3.7 Flash and the GPT-5 Nano variants, useful for short questions and quick lookups. Switch between them at any moment from Vector Config without losing your chat history.

Privacy

All config and code artifacts stay on the device, inside /Vector-AI/ and /Vector-Artifacts/ in Files. No telemetry is collected. Outbound traffic only happens when you actually use Vector: AI prompts go to Pollinations AI, web searches go to the MIKLIUM API, and Python runs inside the MIKLIUM sandbox. Flip Privacy Mode on to block any screen or clipboard reading.

FAQ

Why iOS 17 specifically?

A few of Vector’s tool-calling actions only exist from iOS 17 onward, so anything older trips on the very first turn. If your phone updates, you’re set.

Wait, is this actually free?

It runs on the free Pollinations AI tier by default, no card, no signup. If you outgrow those limits, Vector Config has a slot for your own API key from a paid provider.

Can I use it on a flight with no Wi-Fi?

Not really — every model lives in the cloud and the web search, Python sandbox, and image generation all hit external APIs. Even the local-feeling tools like setting a timer need a model reply to fire them.

Version history

VersionDateChanges
9.3 Jun 2026 Moved a few models to may not work- some models still may not work. We recommend Minimax M3 or Step 3.7 Flash- please change your model to one of these if it’s not working.
8.25 Jun 2026 Bugs fixed

You might also like