Episode 3: Re-launch of AI Assistant Brainbox & Teams, plus new Navigator UI.
Summary: Co-Founders, Rhys and Tom, chat about the re-release of Navigator and Wavebox for Teams, including an update on the team 'managed browser' service that's due out soon. They also discuss the recent re-launch of Brainbox, our OpenAI assistant, explaining how it works with some example skills, and the time-saving benefits it brings to the browser. Plus, they share thoughts on what's to come with GPT-4.
[00:06] Rhys: Welcome to the Wavebox podcast. This podcast is a discussion between the co-founders of Wavebox. Now, Wavebox is a browser for work. It's focused on productivity and it's pretty revolutionary. It's a feature-rich chromium browser, and it's built for productive working across things such as Google Workspaces, Microsoft Teams, Monday, Flashon, Asana, and the list goes on. Pretty much every app you use to get work done will work better in Wavebox. Today it's me and Tom.
[00:39] Tom: Hi there.
[00:40] Rhys: And we've got lots to talk about in this podcast.
[00:44] Tom: Yeah. So we should probably talk about the tab Navigator first. When we first wrote Wavebox years ago, Wavebox was originally based on Electron, and one of the things that we always struggled with Electron was managing memory and resource usage and stuff like that. So we wrote the manager to kind of help with that so you could see visually which tabs and apps were using more memory. We dropped Electron moved to Chromium. These problems have kind of dissipated. However, there's still the problem of how to switch between tabs and apps quickly that's still there. So that's why we've reworked Tab Navigator so it's easier to do this. You can find a Tab Navigator in the bottom left, like a little compass icon - click on that. You can see your apps and tabs in a nice list. You can use the keyboard or the mouse to navigate between them and it just gives you a really quick and easy way to have an overview of everything that's open and kind of switch between things visually.
[01:42] Rhys: Yes. I think the analogy I quite like to use, it's like Google Maps for your tabs!
[01:48] Tom: Yeah, it is.
[01:49] Rhys: So you can get a visual layout of where everything is. It follows the same group structure that you have already on the left-hand side in the webdock in Wavebox. And then the apps and tabs are all laid out in relation to those. And normally, you're zoomed in on the app that you're using or the tab that you're using, but if you want to see what's going on, you can just zoom out.
[02:11] Tom: You can zoom right out. You can literally see every single app and tab that's running. It's like in Google Maps when you zoom all the way out and you're kind of looking at the whole world!
[02:20] Rhys: Brilliant. And it's another way of navigating, isn't it? Because if you are used to doing an Alt or Control tab, depending on your key bindings, that'll take you through the last kind of five or six recent apps or tabs you're using. But if you want a whole picture, click on that, and you'll see everything and you'll be able to navigate either by the mouse or by the keyboard shortcuts. Yeah.
Now, the other big change is the relaunch of teams. We've been working quite hard on this and closely with a few companies who have expanded Wavebox into a group of users or a team of users inside that company and it's now launched for everybody. So if you're a Pro user, you can now have a look at your new management screens. And on there, there is a place where you can actually invite other people. You just put in their email address and they can be invited to join you and use Wavebox. Now, Teams is more than just that because there are some core features in teams that allow you to collaborate together within Wavebox, aren't they?
[03:31] Tom: Yeah, there are. So probably the most obvious one is Connect, which opens up in the sidebar and it gives you a way to chat with your team members. It also gives you a way to screen-share, video, and call share files. And if you've got a team that you work really close with closely, it gives you a quick way to just kind of hop in and out. It's not meant to replace Slack or Microsoft teams or anything like that, but it's that remote worker type thing where you just want to hop on a call and ask a quick question, and then yes, it's.
[04:03] Rhys: Equivalent of, if you're normally sat next to somebody and you would just see if they're deep in work or not, and if they're not, you can just ask them a quick question. Like you say, Zoom or Teams or whatever, you need to schedule a meeting and get involved, and it's a little bit more formal and that normally means you don't bother. But with Connect, because you know exactly who's online and not who's busy or not, you can drop them either a quick note or, like you say, you can have a quick chat with them—you can call them up, do a screen-share. And all of this is one click away, isn't it?
[04:38] Tom: Yeah. And I mean, we use this when some of us are working remotely just to kind of hop in and out if we need to catch up on something.
[04:45] Rhys: Yeah, that's right. And it's a very nice place where you can because if you're doing a screen-share, both of you can screen-share at the same time. So if you're collaborating on something within Wavebox, then the distance between you doesn't become an issue. It's as if you're you're sat next to each other getting the work done.
[05:01] Tom: And in fact, we have used it next to each other, haven't we, for screen-sharing?
[05:05] Rhys: We have indeed. We find that it's easier to have each other's screen on our monitors rather than crouching over somebody looking over at their laptop as they're working on something.
Other things on Teams are helping people to set up. Now, Waybox is very powerful, and sometimes if you come to it kind of unawares, it can be maybe a little bit daunting about, oh, where do I start? Because there are so many things it can do with Teams. You can actually create a template of apps and tabs and a complete setup for Waybox that you can then give to new team users. So this is the Wavebox they get when they download and log in.
[05:44] Tom: Yeah. So it could be maybe everybody on your team has a Gmail account and a Slack. Maybe you use ClickUp for task management and just a few other things. So when a member comes on board, they've got those already, they just need to sign in, and it helps that onboarding process if you've got maybe a medium-sized team to get going. And yeah, it's really powerful for that kind of thing.
[06:13] Rhys: And the work we're doing now will extend that so that you can actually have ongoing managed accounts. What we mean by that is, I can actually control the apps that an individual is using. So if it's a front-of-house kind of place, or somewhere where it needs to be more locked down, then through Wavebox, I can make sure that only the apps that are needed are actually present in Wavebox. And whatever apps I need to add or remove get propagated straight away (managed remotely).
[06:39] Tom: Yeah, we're trialing at the moment.
[06:42] Rhys: By all means, if you're listening to this and thinking now, that would be great in my use case, reach out to us at hello@ or support@, we'll gladly talk to you and see what we can learn together as we trial these features.
[06:57] Tom: Yeah.
[07:02] Rhys: Third thing we want to talk about, which seems to be still everywhere, the....
[07:08] Tom: The Internet is still on fire.
[07:10] Rhys: It does feel like a phase change in what's going on is ChatGPT. It's AI, it's OpenAI, it's basically being able to have a dialogue with your computer that actually is engaging and comprehensive in its understanding of what you're talking about.
[07:28] Tom: Yeah. So we actually started playing with OpenAI and GPT stuff a couple of years ago with something called Brainbox, but it was a little bit too early because you still had to set the scene for the AI. So it's things like you had to create skills and give it examples of how to do things, and then it would do a reasonably good job at copying that, but it meant that you had to preset things up. But the advancement that we've been seeing from OpenAI, and also just some of the changes that we've made internally, means that we can now bring basically just free-form chat. So you can ask the AI, hey, how do I create a recurring event and it will know that you're on Google Calendar, and it just pumps out instructions and it's just really helpful. You used it for something the other day, didn't you? For reformatting some data.
[08:22] Rhys: I did, yes. So I do quite a lot of stuff with some financial information. I want to paste them into a spreadsheet and in the past, I'd have to write a little Python program that would take some data and munge in a certain way so that I could actually then calculate things in Excel. In the recent one, we tried with Brainbox and I literally copied the data I was working off on the page that I was viewing, and I asked Brainbox to summarize the numbers, strip out the $ or £ sign in front of it, and get rid of the commas. So that is the output I had. I literally then just pasted it into the spreadsheet, and it all worked absolutely fine and saved me, what will probably be about 20 minutes of a quick Python code just to do some data cleaning - it's like a second brain. That's why we called it Brainbox. It's there, always waiting to help you. Now, the icon in Brainbox is a little smiley....
[09:26] Tom: It's a purple smiley face.
[09:27] Rhys: You click on that, and on the right-hand side, there will be a dialog view. And this is where you actually interact with Brainbox. You can type any question you like in there. You mentioned that it has context, doesn't it?
[09:41] Tom: It does. So there are three different levels of context. The first one is it's got a basic understanding of what you're viewing. So, for example, it'll know that you're just viewing Google Calendar but won't know anything about what's in Google Calendar. If you want to ask it something about what's on the page, you can ask it to summarize it. But this will need to send the page obviously off to OpenAI. So it's opt-in. There's a little kind of chip that comes up above the text field and then you can tick that to say, yeah, send the page with it. So when you're saying summarize this content, tick that and it sends it off and it will come back with the content of what's there. You can also highlight the text on the page and that's a similar experience. So it's always opt-in in that it will only send stuff along when you ask it to. You can also do some of these things through the right-click menus. So we've got the skills there. So these are really good for summarizing things. Or, if you want to create a reply to an email, you can highlight the email you want to reply to, right-click Brainbox > Draft a Reply. And it's those kinds of things that really kind of power it up.
[10:47] Rhys: Yes. I mean, it's good to note that the way that you guide OpenAI or GPT style language models, in general, is you set the scene for them, and this excites them in a certain direction so that when they actually get the content, they know how to deal with it. So it can be something like you are a bot that can summarize pages of documentation all into one sentence, or given a paragraph, you can create a snappy headline that you can use as an intro. So this basically sets the scene. And normally, if you were typing something in on the right-hand side, you'd have to type this in every single time. It gets quite a bit tiring. What we've done in Wavebox allows you to encapsulate some of these into what we call skills. And these are kind of pre-canned warm-up kind of strings, effectively, that you send to Brainbox to OpenAI, and it tells it what it should be doing with the content that will then follow. And the examples there that you gave were things like being able to reply to an email. So you've highlighted email, you go right-click skills, reply and it knows I meant to. This is a conversation with somebody on email and I've now got the content and I can then generate that. We have allowed people now to create their own skills.
[12:10] Tom: You can. And I think Sue published a blog yesterday. She's got some other skills in there that you can try out.
[12:16] Rhys: Going back to the example I found this week where this thing of taking some financial information from one app in Wavebox and pasting it into another app in Wavebox, I can type into Brainbox. Please strip out blah blah blah for these numbers and give it to in a table. Now, because I do that every month, I can actually create a skill from that next month. All I'll do is I'll select everything on this page, right-click skill, generate the numbers for me, and great, straight away, I can just paste it in where I want to go.
[12:47] Tom: Exactly. So you've saved like 20 minutes a month using Brainbox.
[12:50] Rhys: That's it. And the case here is that it's clever enough to know where to find the data on the page, and it's clever enough to know to change it to a different format. So if you're doing quite a lot of copying and pasting from app to app, I think putting Brainbox in the middle there will help you enormously if you're having to do any kind of manual managing of data. So skills as well are available in the team's setting. So if you create new skills or you have skills that you want to use, that you pick up elsewhere, you can actually have them as part of the team offering. So all the team members can have skills directly themselves. And then in terms of the kind of viewing forward, again, there's lots happening in the industry with language models. I mean, over the next year, shall we say, what does that look like for Brainbox and its progression?
[13:42] Tom: So Brainbox uses something called GPT-3.5. So this is the progression we've gone from 3 to 3.5. OpenAI announced GPT-4, which is again another improvement on that. It accepts things like images as a kind of prompt content, it gives better replies, and stuff like this. They are going to be releasing that to developers shortly. So I mean, this will just be the progression that we follow along as well. We'll be using those things from there to improve it and kind of move along with it. And I think we spoke about giving the option to use either 3.5 or 4, depending on your use case, because 4 as well gives you a much bigger context window, so you can send a lot more to it and get longer replies.
[14:27] Rhys: It does, yes, that's right. So this is the equivalent of, say, your short-term memory with the current GPT. It's around about 4000 tokens. Now, the token is roughly not quite a word.
[14:40] Tom: It's a bit more than a word, isn't it?
[14:42] Rhys: Yeah. So 4000 gives you enough to respond to emails and all the rest of it. But if you were looking at, say, a very large financial report or a Wikipedia article, whatever it is, that is pages, pages long, it doesn't have sufficient token memory to be able to comprehend that in one go. You have to select a section and work on that and then iterate on that as you go through the document with GPT-4, they've enhanced that from 4000 to a maximum of 32,000, a big eight-fold increase there.
[15:13] Tom: It's huge, isn't it?
[15:14] Rhys: And that really does change things that basically you give somebody a thesis, you can paste it in and ask it to critique it or summarize it one paragraph, it's large enough actually to comprehend quite large complex documents. Being able then to combine that with images makes it very, very powerful because if you're using something in Wavebox where maybe there's a scanned image and it's a little diagram, but you want to extract something from it previously, you'd have to recreate that or basically do the manual work itself. Here you can include the image in the query that you send to Brainbox and that can be things like please regenerate this or interpret it given some other kind of data and output it in whatever format I'm asking. It might be text, it might be a table, or it might be another image. So those things we're looking forward to, to enhancing Brainbox with GPT-4 over the course of the next year as it's being released. It's also possible to use other language models. We know that Google has released Palm this week, so again, we'll be looking to integrate all the different language models out there and allowing you as a user of Wavebox to kind of pick and choose which one you want to use depending on the task that you have in hand.
[16:34] Tom: Yeah, exactly, because I think some language models are going to be better than other ones at certain tasks. And the other thing is that we've got a token quota because there's obviously a real-world cost to generating these replies. And some language models work out cheaper and others are a bit more expensive. I think the Wavebox Pro plan gives you 100,000 tokens.
[16:55] Rhys: It is, yeah.
[16:56] Tom: And then you can top that up to two and a half million for an extra $5 a month.
[17:02] Rhys: Yes, that's right. I mean, we're basically projecting existing costs to us and pretty much offering that to you direct. And if you're an avid user, we also allow you actually to enter your own API key.
[17:14] Tom: Yes, that's the OpenAI API key. So you can build directly with OpenAI.
[17:19] Rhys: That's right. In which case, we'll be just a conduit for the OpenAI chat. We allow that relationship directly between OpenAI and yourself. I think this is a good way for people to be able to use Brainbox to its maximum use. Yes, it's there for you all the time. And if you're using Waybox now and you want to give it a go, click that icon at the top. The color is purplish and it's a happy smiley face.
[17:46] Tom: Yes.
[17:47] Rhys: And then type away. And please give us feedback. We are actively developing this. So, again, drop a note to us at hello@waveboxio or support@ and we'd be delighted to hear what you think and help guide us on how we develop Brainbox in the future.
[18:02] Tom: Yeah, that'd be really good to see it go forward.
[18:04] Rhys: Great. Well, I think that's it for the podcast. Lots of things to try, and we'll give you an update next month on the developments in Wavebox.