1LocalAITool v0.1.0 — the first public release

New

• Setup picks the model for you. The app measures this computer, recommends a model that fits its memory, downloads it with a checksum check, and resumes a download that was interrupted.

• A guided first job shows the whole loop — queue it, watch it run, open the answer — before you leave anything overnight.

• Honest speed. Every task shows how long it takes *here*, marked as measured or estimated; one click runs a real benchmark.

• Hand it work and walk away. Summarize, translate into sixteen languages, ask a question, summarize a whole PDF, make a picture, clean up a recording, transcribe audio, read text aloud, build a study course, turn a picture into a 3D model, and fill a form from a document.

• Long documents finish. Big inputs are split, worked through in pieces and stitched back together instead of running out of room.

• You decide when it works: any idle time or set hours, wait for the charger on a laptop, how much of the machine to use, pause everything with one switch. It estimates what a night costs in electricity.

• It survives the night. Closing the window keeps it in the tray, it can start at login, it keeps the machine awake while a job runs, and it recovers the engine if it crashes.

• Routines repeat a job on a schedule and keep a logbook of every run. A machine that slept owes one catch-up run, not a pile.

• The inbox holds finished work and the queue: open, copy, retry, cancel or delete any job.

• Chat on your own hardware, with folders, streamed replies, pictures in and out, and a model chosen for this computer.

• Phone inbox. Scan a code, add the page to your Home Screen, and send work or read answers from your phone. It updates live; the phone never runs the model.

• Household pool. Link another computer with a six-digit code and queued jobs go to whichever machine is awake and able.

• OpenAI- and Anthropic-compatible endpoints on 127.0.0.1, with per-tool keys that are shown once and stored only as a fingerprint.

• MCP tools and the 1lat CLI to queue and read jobs from scripts and agents.

• Code jobs — review a diff, write tests, write docs, draft a commit message — and one-click terminal coding agents wired to the local model.

• Skills a job can use: web lookups and read-only commands (off until you turn them on), slide decks and PDFs as output.

• Everything is free to use. Pro lifts three lifetime caps: 30 overnight jobs, 200 chat messages and 2 hand-picked chat models. Reading old chats and opening finished work is never limited.

• One payment, no subscription. Single Device $29, Three Devices $59, Five Devices $89. Twelve months of updates are included, and the build you have keeps working forever.

• Licenses are verified offline against a signed claim bound to this computer. Offline for a while? A license keeps Pro for a 14-day grace window until the app can check in again.

• Nothing leaves your computer. No account, no upload. The app talks only to the license server, the update feed, and — only if you say yes — an anonymous usage counter that cannot carry your content.

• Self-updating from signed releases, checked at launch and every six hours while the app runs.

• macOS 13 or later on Apple Silicon and Intel; Windows 10 and 11 (64-bit). The Windows installer is not code-signed yet, so SmartScreen may ask you to confirm the first run.