You already own that chalk line — you just can’t find it, so you buy another. Tool Vision labels and files every tool by camera and voice, so putting things away and cataloging them become one two-second motion.
So the garage stays chaos — and chaos quietly costs money every weekend.
That’s the whole loop. The camera watches the bin; you just keep putting tools away.
Later — ask “where’s my chalk line?” and it tells you the bin.
Started in an AI app builder, went quiet, then came back as a full hands-free system. 205 commits, front to back.
Two AI eras in one repo — generate the scaffold (2025), agentically build the hard system (2026).
Local-first, open models, no vendor lock-in. Every layer swappable.
Not a roadmap. Every one of these ships right now.
Tool inventory today is either enterprise asset-tracking (priced per seat) or generic label apps that still make you type. Nothing is hands-free, local, and free.
| Capability | Tool Vision | Sortly | Brother P-touch | Spreadsheet |
|---|
The first hands-free, local, open-source tool inventory — priced at zero because the moat is the workflow, not a subscription.
Tool Vision ships an official MCP connector — Claude creates bins, catalogs tools, sets categories, and prints labels headlessly. No browser needed.
list_locationscreate_locationcreate_bin_walldelete_locationset_bin_categorylist_binsbin_contentslist_itemscreate_itemplace_itemmint_short_codecapture_webcamrequest_phone_photolist_captured_photosupload_captureprint_labelMost AI lives behind a keyboard. The real leverage is putting it where hands are busy and screens are inconvenient — the workbench, the job site, the shop floor.
Tradespeople, mechanics, makers, and DIYers lose real hours and money to disorganization. The interface that fits that world isn’t a form — it’s a camera and a voice. Tool Vision proves agentic AI can do useful physical-world work: see an object, understand it, act — print, file, retrieve.
The pattern generalizes: parts rooms, warehouses, lab inventory, job-site materials. Everything has a spot; the AI keeps it that way.
Speech runs on whisper.cpp on-device — no cloud microphone, no per-word billing, nothing leaves the garage. Vision goes through a swappable, bring-your-own-key Worker, so you choose the model. The label logic, the print queue, the agent tools — all open source (MIT).
Capable enough to be useful, open enough to trust, cheap enough ($0) that a hobbyist and a shop both just run it.
Every number on this page is read live from the same Supabase the app writes to — my own inventory, not seed data.
The dense one. Bins like USB-C to USB-C Cables — B7 and Digital Optical Cables — B15 exist because the alternative was a drawer I dug through every time.
These are frames Rapid Mode grabbed off the webcam in my garage. The vision model named each one, I said “yes”, and the label printed. No stock photography anywhere on this page.
The app itself, on this inventory. These four panels are interface recreations — rebuilt from the product’s own design tokens and populated with the live numbers above, not captured off a screen.
The animated walkthrough runs the whole loop — open a bin, present a tool, say yes, watch the label print.