A suggestion that I think can provide a lot of value to LAM/Teach Mode/Playground:
For WEB LAM:
-
Ability for a user to save a template VM instance and/or browser configuration and/or keychain token that (Like with BRAVE browser BraveSync) that would enable LAM to save and easily import a) browser extensions, and b) bookmarks into each unique LAM user’s session(s), across sessions.
-
if possible, support saving the cache of each extension for users, for instance, it a user has METAMASK installed as a browser extension, allow the configuration file(s) to be stored in the same way that session credentials currently are — protected, but easily importable and available for a VM on-demand. This would support the ability for LAM to access crypto wallets.
— but also could provide additional functionalities provided by popular features. -
With the bookmark sync noted above, let LAM first look through it’s bookmarks for a source page before searching the web — maybe having direct access to a JSON version of the bookmarks (and the folder structure it can see on screen, so it know what folder has what link), using weblinks it knows as TRUSTED SOURCES (based on the user) in real-time, rather than always having to Google/Search everything — the user could say “GO TO OPENSEA AND FIND CRYPTOPUNKS”, and instead of googling OpenSea Cryptopunks, it would go directly to opensea, then search the searchbar for cryptopunks. This reduces the need for LAM to rely on google and/or perplexity for search, and instead allows users to provide a sort of KNOWLEDGE BASE of sources or webapps.
Worst thing we’d want is someone saying “go send a message on XXX to XXXX”, and the LAM searches for the site, only for it to click a fake promoted site, or scam, and then getting stuck when it’s not logged in already.
———
The base rationale for this functionality would both enable crypto transactions using R1 in a more secure way from hot wallets, additionally, a user could provide R1-specific accounts to LAM/rabbit for use across various services. For instance, I run a discord server, and having the ability for LAM to “go to ABC site, find the latest post from X user, draft XYZ based on [context], and DM it to me, [username] on Discord. After I edit the draft, I can say, “go log onto Twitter and create a draft article of XYZ using the response sent to you on discord from X”
I know this isn’t necessarily anything game changing from what we can do now — but being able to inject trusted sites, or, sites/bookmarks to prioritize when going to do an action for an item, could really add a lot of functionality.
“Go through my News bookmarks and compile a daily update for me on the state of global affairs using sources using all listed publications, and send me a message on Discord”.
While we’re at it — why not create the ability to create a contacts list that can put a name on the user “You” (on iPhone), and then a list of socials for all people you might interact with. This would provide a ton of value for interactions and confirming that the right user is getting the message you want, rather than leaving it up to chance (such as recognizing their username or not directing the user/interacting with a fake/scam account that you might have not previously interacted with, or interacted with in a while.)