OK, iāll try and iāll let you know. The only time I tried this it did sign me in, but at the end of the session, asking my R1 to perform a task I previously signed into Rabbit Hole, it forgot the credentials.Are you prompting to sign in via the web interface on the rabbit hole? Or you dictate your credentials via Voice to R1?
Could you send me those prompts? Iāll be happy to try out.
1 Like
Iām really excited about this great feature!
In the Rabbit hole, if we log in to our account on Amazon for example, we donāt seem to stay logged in afterwards. So when I ask to add a product to my cart, I canāt find it on my Amazon account via another device or the Amazon app for example.
Would it be possible to keep accounts permanently connected (until we log out ourselves)?
2 Likes
I can confirm login Cookies are stored once you log in and the task goes successful. I connected R1 to my Facebook, Instagram and Google accounts and it performs wonderfully!
4 Likes
Great job team Rabbit!
The LAM playground is so addictive. It feels like my R1 is on nitro mode that it can do a bunch of magic with just a few simple but clear instructions. Keep up the good work guys. Lovināit!
4 Likes
I like writing longer prompts for the (web) LAM playground and it would be great if the input field would grow to show the whole prompt.
4 Likes
Great suggestion! Expanding the input field for longer prompts would definitely enhance usability.
[Written by R1]
3 Likes
Instead of reinventing the wheel why donāt you guys may a deal with Multion? It seems like theyāve already built software that does what you are trying to do and it does it lighting quick!
2 Likes
Because we think our approach is better, and theyāre not targeting the consumer market, or making hardware.
4 Likes
2 Likes
Yeah, sometimes it sticks there and wonāt go onā¦ especially in complex websitesā¦
1 Like
Hi Simon,
Where is the best place to report experience of LAM Playground that will be most helpful to the team to make it better?
Is this a good place, or should we do support tickets?
Regards,
Trevor
1 Like
this is amazing! <3 I have been a little bit busy with work lately and didnāt have a time to catch-up with that update yet, but Iām really impressed. one of the bits Iām enjoying so far is simply asking for figuring out the latest deals on specific videogame stores 
looking forward to read more use casesā¦
3 Likes
@simon Does the LAM only work with visual information from the interface? It gets stuck when things are not labeled (for example the email body section in gmail)
Is there a possible fix by also looking at the code of the page?
Itās also trying to do the same actions for like 30 times when stuck. (It might help if it understands what it is trying isnāt working)
2 Likes
I would like a remove/undo button. Sometimes when I try to redo or try to change a prompt, it is stuck on the previous one. An undo/memory wipe of a prompt would be nice.
2 Likes
Hello everyone
2 Likes
If you press 5 volte the PTT Button, youāll refresh itās memory and youāll stop the session. Then you can start over.
1 Like
It does receive the code in Input, as explained in the engineering post. Otherwise it canāt generate the correct script if only receives visual interface information. The fact is that not always things are labeled in code, for example: the body section in GMail is not properly marked in the source of the pace, is also wrapped in an overlay which makes LAM struggle, and it canāt understand the thing.
The fact that it repeats the same action different times is normal when the website is totally new for it. For example: in my case, the first time I logged in to Facebook with the LAM, it had to locate different times the notifications button. The second time I head to the website, it matched the button in a second. And from that time on it never failed.
Rabbit explains it when they talk about Planning, the fase in wich LAM learns how the UI is structured. Since it does not affect the whole training of the model, but only affects the device of the user, each new R1 has to plan the interface every time it gets into a new, unknown website.
@Simon I hope I got it wrightā¦
2 Likes
Thanks for your message. Where can I find the engineering note you are talking about? (I found it, reading it now)
Our latest progress in LAM playground employs an approach that directly interprets your request and will iterate over its planned steps to complete the action. The current approach makes use of both visual input (i.e. the screenshot) and structured input (i.e. the page source code) in order to take actions via a webdriver (e.g. Playwright)
I understand it might try the same thing multiple times but it gets stuck in a loop without anything changingā¦ it would be great if if it would take an alternative approach when things donāt change after many of the exact same actions.
2 Likes
It might be nice to have a more consistent way to start a playground session. Maybe something with a keyword like ābrowseā.
It would also be nice to see if there is a session active in the rabbit itself.
Sometimes the regular Rabbit assistant denies it can browse websites and thatās not great after you described an elaborate action.
1 Like