Ottomate suggests words to complete your sentences as you type, whatever app you're typing in. Hit the Tab key to accept a suggestion. Suggestions become more frequent and relevant over time. All local! We do not send anything you type outside your computer.
* We’re big fans of Gmail SmartCompose but sadly it’s just for Gmail and Docs, so we decided to try and make it work for any app on a Mac.
* We hacked together some version of GPT2 that works locally and everywhere on your Mac. The current model is a bit crude (trained on reddit data) but works.
* We added a very simple n-gram model trained with your own inputs to add some personalization.
* This is a very simple version of the app but it felt good enough we wanted to share it, we’re eager to know what you think of it!
* In the future, we want to improve the main model for everyone. More languages and probably some specialized models based on your usage (chat, email, document editing, programming…)
* We also need to generally improve the app, faster inference speed, smaller binaries… For this we are strongly considering rust which is already used for our very small backend.
* My personal favorite: having it autocomplete "import ip → db;ipdb.set_trace + Esc + :wq" felt sooo awesome. For the non-devs, this inserts a debug line then saves and quits the file. I tend to write this quite a bit when debugging a program.
Very cool, been using it over the past few weeks and it's definitely a plus to have the autocomplete outside of Gmail. How do you plan to manage the “conflicts” within Gmail BTW? Any shortcut to disable Otto temporarily when we don't want to use it?
@robi thanks! and great idea :) We'd like to give you the option to pause it directly from the Menu Bar, or even disable it automatically whenever it's not needed/wanted.
Wonderful app. Do you plan on adding Turkish support?
I always wanted to make an app in python so that it will complete sentences based on app, recipient, time, location and context but I thought required processing power for such a task will cripple most devices. And I realized python is not a good choice for this aim.
Do you think such deep and instant customization possible?
@hsnkts We definitely want to add multiple languages, No definite list yet, as it depends on the kind of dataset we can create to have good models.
As for your second question, we have multiple ideas to reduce deep learning memory footprint. I was pretty surprised, but pretty naive and quite large models seem to work reasonably well and don't seem to cripple machines too much. There is still a large margin of improvement, and is at the top of our todo list at the moment.
Ottomate
Jitter
Ottomate
Jitter
PhotoRoom
Ottomate