Manus AI – Complete Course for Developers

[00:00] In this course, we're going to go over [00:01] everything you need to know about Manis [00:04] AI. By the end of this, you'll know how [00:06] to use it for research, building [00:08] websites, creating presentations, [00:10] automating workflows, and a lot more. [00:13] I'm Bo KS, and I'll be teaching this [00:16] course. Manis provided a grant that made [00:18] this course possible. Let's talk about [00:21] what Manis actually is because it's not [00:24] what most people think. Most AI tools [00:27] you've probably used are chat bots. You [00:29] ask a question, you get a text response. [00:32] That's it. Chat GPT, Gemini Cloud, they [00:34] all work this way at their core. You [00:36] type something, they write something [00:38] back. Manis is different. It's what's [00:41] called an AI agent. Instead of just [00:44] answering answering your questions, it [00:46] actually does things. It can browse the [00:49] web. It can write code. It can create [00:51] files. It can fill out forms on [00:53] websites. It can run tasks on its own, [00:55] step by step, without you babysitting [00:58] every part of it. The team behind Manis [01:01] calls it an action engine. That's a [01:04] pretty good description. You tell it [01:05] what you want done, and it figures out [01:07] how to do it. It breaks your request [01:09] into smaller steps, executes each one, [01:12] and gives you the finished result. [01:14] Here's a simple example. If you ask a [01:16] chatbot to find the top 10 competitors [01:18] in the organic skin car skincare market [01:21] and compare their pricing, it'll give [01:23] you a text response based on whatever it [01:25] has in its training data. Some of the [01:28] info might be outdated, some might be [01:30] made up. If you give that same task to [01:32] Manis, it actually opens a web browser, [01:35] visits real websites, looks up real [01:38] pricing, and compiles everything into a [01:40] spreadsheet or report, and then hands [01:42] you the file. You can watch all do all [01:45] this in real time. That's a [01:47] fundamentally different thing. The [01:49] important thing is what it can do for [01:50] you. And here's one more technical thing [01:52] worth knowing. Every task you give Manis [01:55] runs inside its own isolated cloud [01:58] computer. Think of it like a virtual [02:00] machine that spins up just for your [02:02] task. It has a web browser, a file [02:05] system, networking, the whole thing. [02:07] This is called the sandbox. The sandbox [02:10] setup is actually important because it [02:12] means Madness can do things that regular [02:14] chat bots simply cannot do. It can [02:17] install software, run code, save files, [02:20] access the internet in real time. It's [02:22] working in a real computer environment, [02:24] not just generating text. You can use [02:27] Manis on the web at manis.im. [02:30] There's also a desktop app for Mac and [02:32] Windows and mobile apps for iOS and [02:34] Android. We'll be using the web app for [02:37] most of this tutorial since that's where [02:38] most people start. All right, that's the [02:41] big picture. Now, let's set it up and [02:44] actually use it. [02:48] Okay, head over to manis.im [02:51] and create an account. We can sign up or [02:55] sign in. I'll just sign up. And you can [02:58] sign up with a bunch of different ways. [03:00] I'll just use my Google account. And [03:02] then you can upgrade to manage pro or [03:04] you can just skip it if you want to stay [03:06] on the free plan. And it's available to [03:08] download for all the main operating [03:10] systems or you can just use on the web. [03:13] Once you're in, you'll land on the main [03:15] dashboard. The first thing you'll notice [03:17] is this big input box in the center of [03:20] the screen. It says, "What can I do for [03:22] you?" Below it, you'll typically see [03:25] different things like these suggested [03:26] categories down here like create slides, [03:29] build website, develop desktop apps. [03:31] This input box is where everything [03:33] starts. You basically tell what you want [03:36] man to do. Hit enter and it gets to [03:38] work. And you should be able to follow [03:40] along with this tutorial using the free [03:42] tier. You can do quite a lot with it. [03:44] You can check the link in the [03:45] description for a link to sign up to [03:47] Manis. Now, let's run our first task. [03:49] So, I'm going to say research the top [03:50] five project management tools in 2026 [03:52] and create a comparison table with [03:54] pricing, key features, pros, and cons. [03:56] And then I'll just hit send here. Then [03:58] we'll watch what happens. Now, it's not [04:01] we're not just going to get only text. [04:02] We're going to have this sidebar that's [04:04] going to open on the on the left showing [04:08] us a live activity log. So, it's going [04:11] to show the different tasks that it's [04:13] working on. And you can see it's now [04:16] starting to do the research here. And we [04:18] can see down here we can see um what's [04:22] what it's the task that it's working on. [04:24] And you can even see the browser because [04:26] Manis is using an actual browser and now [04:29] it's using a computer. So if I click [04:31] here, we can actually see the the [04:34] computer that Manis is working on. It's [04:36] using an editor and like I said before [04:39] it was using the browser and you can [04:40] basically see what Manis can see. Here's [04:43] the browser it's actually accessing. So [04:46] we can just watch it work or we don't [04:49] really have to watch. But I think it's [04:51] pretty cool that you can actually watch [04:53] the AI work. There's a live cloud [04:56] browser view that shows you the actual [04:58] web page it's visiting. You can see the [05:01] cursor move. You can see it scroll down [05:03] pages. You can see it click on links. [05:06] It's like watching someone do research [05:08] for you except it's h happening at AI [05:11] speed. And like I said, you don't have [05:13] to watch while it's working. You can [05:14] leave the page. Um we could go grab [05:16] coffee, do something else. the task just [05:18] keeps running in the background on [05:20] Manis' servers. You don't need to even [05:23] keep the tab open. So, I'm just going to [05:26] come back when this is finished. [05:28] Okay. And it's finished. So, we get this [05:33] nice comparison table right here. Plus, [05:37] it created a an interactive website. So, [05:40] I can just click that and see the [05:42] website it created. So, [05:44] this is actually pretty nice. It it [05:47] created this entire website based on the [05:49] prompt and it's giving us a detailed [05:52] tool profiles of all the different [05:54] tools, [05:56] pros and cons. We got a visual [05:59] comparison, starting price comparison. [06:02] So, these are some pretty nice graphs [06:03] that makes us understand a lot better [06:05] pricing breakdown and then a full table. [06:09] In the chat, we just had a simplified [06:11] table, but here's the full table with a [06:13] lot more information. And then it's even [06:15] giving us a final verdict of helping us [06:18] find out what tool is best for us and a [06:20] key takeaway. And then we have these [06:23] links to quickly go. So this is a really [06:27] really easy to understand good overview [06:30] of all these different tools. And if we [06:33] wanted, we could create this into a [06:35] permanent website, but I'm not going to [06:37] do that. We can also view all the files [06:40] within this task. So here's all the [06:41] files. We could download these files and [06:45] including the code files and basically [06:48] continue to work on the site or do [06:50] whatever we wanted with it. And if we [06:52] view Manis's computer like this, we can [06:55] see the computer right within um loaded [06:59] on we can see the website loaded on [07:01] Manis's computer. One thing to keep in [07:04] mind is that Manis tasks can take a few [07:07] minutes, especially if they involve a [07:09] lot of web browsing. that that's normal. [07:12] It's actually doing real work. It's [07:14] visiting real websites. It's reading [07:15] real content. It's not just pulling from [07:18] a stack database or just uh getting the [07:21] information from the model weights or [07:22] anything. Uh the trade-off for getting [07:24] real current information is that it can [07:27] take a bit longer than a chatbot [07:29] response. Okay. So, now let's talk about [07:32] something very important. How to write [07:34] good prompts for manis. This is honestly [07:37] the most important skill you'll develop, [07:39] and it's different from prompting a [07:41] chatbot. With a chatbot, you ask a [07:44] question and expect a reply. With Manis, [07:47] you're giving it a job to do. Think of [07:49] it like delegating to a really capable [07:51] assistant. You wouldn't just say [07:54] marketing to your assistant. You'd say, [07:56] "I need you to research our three [07:58] biggest competitors, pull their latest [08:00] pricing, and put it in a spreadsheet by [08:02] Friday." [08:03] The same logic applies to Manis. Be [08:06] specific about what you want. Tell it [08:09] what format you want the output in. Give [08:11] it enough context about why you need it. [08:14] They also recommend a template pattern [08:16] where you use brackets for [08:18] customization. For example, create a [08:20] competitive analysis of industry [08:23] focusing on metric one and metric 2 [08:26] presented as a format. Here's a good [08:29] universal starter prompt if you're not [08:31] sure how to frame something. I want you [08:34] to take ownership of helping me with [08:36] your situation. Figure out what needs to [08:39] be done, ask me questions if needed, and [08:41] manage the process. That kind of prompt [08:44] gives Mannis the freedom to figure out [08:47] the right approach, which is often [08:50] better than trying to micromanage every [08:53] step. A few more tips on prompting. If [08:56] you have existing files that are [08:58] relevant, upload them. You can drag and [09:00] drop files directly onto the chat. Manis [09:03] can read PDFs, spreadsheets, word [09:04] documents, images, all sorts of things. [09:07] Instead of describing your data from [09:09] scratch, just upload the file and say, [09:11] "Restructure this into a clean report or [09:14] analyze this spreadsheet and give me the [09:16] key insights." Also, be explicit about [09:19] what format you want. Give me a PDF [09:22] report. Put this in a spreadsheet. [09:24] Create a presentation with 10 slides. [09:27] The clearer you are, the better the [09:28] output. And one more thing, Manis has a [09:31] few different modes. Chat mode is for [09:34] quick questions that don't need [09:36] multi-step execution. Agent mode is what [09:39] kicks in when you give it a real task. [09:41] The system handles this automatically [09:44] based on your prompt. But if you notice [09:46] Mannis just giving you a text reply when [09:49] you wanted it to actually do something, [09:52] try rephrasing your prompt to be more [09:54] taskoriented. [09:59] Research is probably the single most [10:01] common use case for Manis, and it's [10:03] where the difference between an agent [10:06] and a chatbot becomes really obvious. [10:09] Okay, let's start with some basic [10:11] research. I'm going to give Manis a [10:13] research task and walk through what [10:16] happens. So, I'm putting in research the [10:18] current state of the electric vehicle [10:20] market in Europe. Cover the top selling [10:22] models, market share by manufacturer, [10:23] government incentives by country, and [10:24] upcoming models for the 2026, 2027. Put [10:27] everything into a well organized report. [10:30] Okay? And it's going to start thinking. [10:31] It's going to start working. Now, the [10:33] previous example also involved research, [10:35] but this is arguably more research. So, [10:39] we can see this the task progress down [10:42] here, all the tasks that it's working [10:44] on. We can open up the browser and then [10:48] we can see the what it's doing in the [10:50] browser just like before. So, it's going [10:52] to decide to search for several [10:54] different things. It's going to look up [10:56] the EV sales data, manufacturer market [10:58] share, government incentive programs, [11:00] and more. And it's going to have to [11:02] visit multiple sources for each topic to [11:05] cross reference information. We'll see [11:07] it browsing news sites, industry [11:09] reports, government websites, automotive [11:12] publications. [11:13] It's basically going to read through [11:15] each page, extract the relevant [11:17] information, and compile it all [11:20] together. [11:21] So, this would take hours to do [11:24] manually. Manis is going to do it in [11:27] just a few minutes. I once it finishes, [11:30] we'll we'll kind of see how it turns [11:31] out. But I also want to talk to you [11:33] about wide research. This is one of [11:36] Manis's most powerful features, and it's [11:39] specifically designed for tasks that [11:41] involve researching many items. So, [11:44] here's the problem it solves. When a [11:46] regular AI agent researches items one at [11:49] a time in sequence, the quality can [11:52] sometimes start to drop after about [11:53] eight items. The context window fills [11:56] up. The model starts cutting corners, [11:59] and then by item 20 or 30, you're [12:02] getting vague or even fabricated [12:04] information. [12:05] This is a real technical limitation. [12:08] Wide research fixes this by spinning up [12:12] multiple manis instances at the same [12:14] time, each in its own fresh sandbox. [12:19] So, for example, if you ask it to [12:21] research 50 companies, it doesn't [12:24] research them one by one. It sends out [12:27] parallel agents each with a clean [12:30] context to handle subsets of the work [12:33] simultaneously. [12:35] So item 50 gets the same quality of [12:38] research as item one. [12:40] This feature kicks in automatically when [12:43] Manis detects that your task involves [12:47] many similar items. You don't need to [12:49] enable it manually, but it's good to [12:52] know when to use it. Wide research is [12:55] great for things like researching a long [12:58] list of companies, doing a comprehensive [13:01] literature review, comparing many [13:04] products or services, or processing a [13:07] large data set where each item needs [13:09] individual attention. It's not as useful [13:13] for tasks with fewer than 10 items or [13:15] for deeply sequential work where each [13:17] step depends on the previous one. For [13:20] those, regular research works fine. Now, [13:23] it looks like we were just It looks like [13:25] this task finished. And let's see what [13:28] it came up with. Basically, there's a [13:32] PDF version and a markdown version. So, [13:34] here's the markdown version with all the [13:37] information it came up with. And so, [13:40] that's some pretty information. We also [13:42] have the PDF version. And we can see [13:46] it's just like the markdown version, but [13:48] it's looks nicer than the PDF. Now, this [13:51] time I don't think it needed to use the [13:54] wide research that I was just talking [13:56] about, but let's try another search [13:59] where it may have to have the wide [14:02] research uh kick in. So, I'm going to [14:05] close this and we'll do a new task. And [14:09] this I'm going to say research the top [14:11] 30 SAS companies by revenue in 2025 and [14:14] then find a bunch of information about [14:16] it and organize them into a [14:19] comprehensive spreadsheet. Now we're [14:21] going to start this and we're just going [14:24] to just watch the log and see if it uses [14:28] wide research. So first it's going to [14:30] get the list of the top 30 SAS companies [14:33] and you can see when it's thinking it [14:34] can even so it ran a Python script. So [14:36] it actually can run scripts to figure [14:38] out things. And it looks like that task [14:41] was actually too easy for manis that [14:44] didn't even need wide research and was [14:46] able to give us a detailed spreadsheet [14:48] with accurate consistent data for all 30 [14:51] companies. But now I want to talk to you [14:54] about browser operator. This is [14:57] particularly useful for a specific [14:59] problem. Basically research that [15:01] requires logging into websites. When [15:04] Manis does research in a cloud sandbox, [15:07] it's browsing the web from a data center [15:10] IP address. It doesn't have your login [15:12] credentials. So, pay well sites, [15:15] subscription databases, your company [15:17] CRM, anything that requires login is off [15:19] limits. Browser operator solves this. [15:22] It's a Chrome or Edge browser extension [15:25] that lets Manis operate inside your [15:27] local browser. Your local browser [15:29] already has your saved passwords, your [15:31] active sessions, and your cookies. So [15:34] manis can access sites when you've [15:37] already logged in. So let me show you [15:39] how to set that up. I'm going to go down [15:41] here to settings here and I'm going to [15:44] go over to connectors and I'm going to [15:46] click add connectors. [15:48] Now there are a bunch of different [15:51] things we could connect Manis to. [15:53] Basically, a lot of the different [15:55] websites that you use, you can connect [15:57] to Manis like Stripe, GitHub, Slack, [16:00] Instagram, Google, Google Drive, Gmail. [16:04] We're going to click on my browser and [16:07] this is going to allow us to access my [16:10] browser. So, I'll just click the connect [16:12] button here and then install extension. [16:15] So, it's going to open up into a new tab [16:17] and then I just have to click add to [16:20] Chrome and add extension and it's going [16:23] to show us how to use it here. Now, if [16:26] you see what it says right here, close [16:29] sensitive tabs to revoke off. So, it's [16:32] only going to have the authentication if [16:34] your tabs are open to that website. So, [16:38] I'm just going to click try it out. So, [16:40] it automatically put this example to try [16:43] it out. This is the YouTube news [16:45] channel. So if I send that, we'll see [16:48] what happens. And now I can authorize a [16:51] new tab in your browser. So I'm going to [16:54] click authorize. And now we can see that [16:57] it has opened up this tab. Manis is [17:00] actually controlling this tab right [17:03] here, right within my browser. So it [17:06] says, see it says Manis is browsing. And [17:09] you can see we're logged in to the Free [17:12] Co camp YouTube channel. So, we need to [17:15] be very careful with this and you need [17:17] to be very careful about what tabs that [17:20] you allow manage to control uh because [17:23] they it can do some real damage. And you [17:26] can see in this top left corner, it's [17:28] going to also show what it's what task [17:31] it's working on. And we we can see that [17:34] it's actually completed the tasks. So, [17:37] it gave us this information. It's ex [17:40] it's explored the YouTube news page and [17:43] it's gotten all this information about [17:44] different news stories that have [17:46] happened. And then we even have a text [17:49] overview of this. And one thing to know [17:52] about the browser operator is that you [17:54] maintain control the entire time. If man [17:58] is Manis is on a page and you want to [18:00] take over, just click inside the tab. If [18:02] you want to to stop it completely, close [18:04] the tab. This is really useful for [18:07] accessing things like similar web, [18:09] crunch base, pitchbook, financial times, [18:12] semrush or any other subscription [18:14] service you pay for. Instead of copy [18:16] pasting data from those sites yourself, [18:19] man can pull all the information [18:21] directly. Now, one thing you should know [18:23] is browser operator can struggle with [18:24] some complex interactions. Drag and drop [18:28] doesn't always work. Multi-step forms [18:29] can be unreliable and you should be [18:31] thoughtful about which sites you [18:32] authorize it to access, especially if [18:34] they contain sensitive data. Just use [18:37] your best judgment. Okay, now let's talk [18:39] about working with files during [18:41] research. You can actually upload files [18:43] directly into a Manis task. You can drag [18:46] a PDF, Excel file, word document, or CF [18:48] CSV into the chat and manis can read and [18:50] process it. So, I'm going to upload a [18:53] PDF right into the chat. This is [18:56] basically a report based on a JavaScript [18:59] curriculum survey from free code camp. [19:02] And it's going to even give me some [19:05] suggested things that we can do with [19:07] Manis. So I'm going to say generate I'm [19:10] just going to click here. Generate a [19:11] presentation summarizing the February [19:13] 2026 survey progress levels. So those [19:16] suggestions were based on the actual [19:18] document that I gave it. So I'm just [19:20] going to submit this. So being able to [19:23] use files like this is great for [19:25] situations where you already have some [19:26] data and need it analyzed or if you need [19:30] it restructured or enriched. For [19:32] example, you could upload a spreadsheet [19:34] of 100 leads and ask Manis to research [19:38] each company and add columns for [19:41] revenue, employee count, and recent [19:44] news. Or you could upload a 50-page PDF [19:47] report and ask for a summary focused on [19:50] specific topics. The combination of file [19:54] uploads and web research is really [19:57] powerful. You can give Manis your [19:59] existing data as a starting point and [20:02] have it go to the web to fill in the [20:04] gaps. And another thing on research is [20:08] that you should treat Manis outputs as a [20:11] really solid first draft, not the final [20:16] product. It does real research on real [20:18] websites, which is much better than a [20:21] chatbot pulling from training data, but [20:24] it can still make mistakes. It might [20:27] misread a number from a website or miss [20:29] a source. It might occasionally get [20:31] something wrong. So, always review the [20:34] output, especially for anything you're [20:36] going to present to clients or [20:37] stakeholders. Think of it as having a [20:40] research assistant who does 90% of the [20:42] work and you get to do the final quality [20:44] check. Okay. And this task is complete. [20:49] Let's check this out. So, JavaScript [20:51] survey curriculum. Oh, this is the [20:52] actual one I gave it. But I can open [20:55] this and open up the slides. So this [21:00] like all these diagrams are not from the [21:02] actual PDF. The PDF just has a bunch of [21:05] data. And so this is a really nice [21:08] breakdown of everything. And wow, this [21:12] is this is really nice. I was able to [21:14] extract all the data from the PDF and [21:17] then make these diagrams. So, I can even [21:20] press play to go into presenter view. [21:24] And I can just go through the slideshow [21:27] right from Manis like this. And if I [21:30] don't want to do the slideshow from [21:33] Manis, I can go to download. I can [21:36] download as a a PowerPoint, a PDF. I can [21:39] convert to Google Slides. So, there's a [21:42] lot of different options of how I can [21:44] access this information. Plus, I can [21:46] actually edit this slideshow right from [21:49] Manis. See, I can move this around. I [21:52] can change this text here. And then I [21:55] can even add speaker notes. And then I [21:58] can just download with the updates that [22:00] I've uh that I've added. And I'm going [22:03] to click these three little dots. I'm [22:05] going to go to code here. Now we can [22:07] actually look at the code that was used [22:11] to create this slideshow or to create [22:14] this slide. So each slide is actually [22:17] created by code which can be easily [22:20] edited. Another thing we can do is click [22:24] this button right here to edit the [22:26] slides. I can describe edits for the [22:29] current slides or I can refresh the [22:32] layout and I can uh change the theme. [22:35] There's a lot of things I can do. So, I [22:37] can go to any of these templates and to [22:39] use these, you have to use Nano Banana [22:41] Pro. This is going to help us design [22:43] prolevel slides with superior layouts [22:46] and basically using Nano Banana Pro to [22:50] help design the slides. And not only can [22:53] it just create slides from this data [22:55] that we gave it, and by the way, here's [22:57] what the the data looked like. See how [22:59] it's all just there's no graphics or [23:01] anything. So it it Manis came up with [23:05] all those charts and diagrams. [23:10] Let's move on to one of man m manis's [23:12] most popular features creating [23:14] presentations and visual content. I [23:17] already showed it a bit in the last [23:18] section, but I want to talk more about [23:20] presentations. So I can just say [23:23] something like create a 12 slide [23:24] investor slide pitch deck for a fintex [23:27] startup called Payflow. and it has all [23:29] the information that I want in the the [23:32] slide deck. And it could even do more [23:34] research. It's not just it's not just [23:36] based on the what we give it. And I'm [23:39] going to click create slides so it knows [23:42] it's for sure going to create slides. [23:44] And we can also turn on Nano Banana Pro [23:47] if we want to create imagebased slides. [23:51] So I'll just send that and it's going to [23:52] get to work. Okay, it finished. It [23:54] created the slides just like before. And [23:58] this is like the layout and colors and [24:02] design looks really cool and nice. But [24:06] I'm going to go even more into the [24:08] editing feature. So if I'll click over [24:10] here, we can design the or describe the [24:13] edits. Add an emoji to represent each [24:16] item on the list. So I'll just submit [24:18] that and it's going to start thinking [24:20] and it's going to update. And you can [24:21] see it's optimizing the competition.html [24:24] HTML because like I said, this is all [24:26] code. So if you don't want the AI to [24:29] edit, you can always download the actual [24:31] code, the HTML, and just edit the the [24:34] actual code. Basically, everything is [24:37] very editable. So if I go into this now, [24:40] we can see an emoji for each one. And I [24:43] could describe a bunch of different [24:44] things. Um, I don't just have to just I [24:47] could describe the diagrams. I could add [24:50] I could edit all slides at once. So it [24:53] just makes it really easy to edit these [24:55] slides. Now I want to talk about design [24:58] view and design mode. So this is [25:00] basically an image editing and [25:02] generation tool built into Manis. The [25:06] key feature here is the mark tool. It [25:08] lets you select specific regions of an [25:10] image and give instructions for just [25:12] that area. So if manage generates an [25:14] image and the background looks great but [25:15] the text in the corner is wrong. You can [25:18] circle the corner and say change this [25:19] text to say Q3 revenue without [25:22] regenerating the whole image. So you can [25:24] mark multiple areas and give separate [25:26] instructions for each one. Or also you [25:28] can use the text editing mode when you [25:29] just need to fix or add text. So let me [25:31] try a design task here. I'm going to [25:33] click the design button here and I'm [25:35] going to say design a social media [25:37] banner for Instagram for a coffee shop. [25:39] Use warm colors, include the the shot [25:41] name prominently, and add the tagline, [25:43] "Your day starts here." So, I can either [25:46] do GPD image 2.0 or Nana Banana Pro. So, [25:50] I'm just going to send that, and it will [25:52] just start. Now, this feature is very [25:55] useful when you're working on design uh [25:58] designs iteratively. You can generate an [26:01] initial image, then refine specific [26:03] parts through targeted edits rather than [26:07] starting over each time. And so right [26:09] now it's currently generating the image. [26:12] Okay. And here is the image. This looks [26:15] pretty nice. And there's a few ways that [26:18] I can actually edit this. So one is the [26:21] comment tool. So I'm going to click [26:23] comment tool. I'm going to select this [26:24] and I'm going to say make a greener [26:27] color. And then I can go over to um the [26:32] select tool. And if I click here, it [26:35] kind of to zoom out a little bit. I have [26:37] upscale and remove background or edit [26:39] text. I'm going to click edit text and [26:41] it says your day starts here and I'm [26:44] just going to say the way to start your [26:47] day and I'll click save. So now it has a [26:51] few things that it can edit and it's [26:53] going to do one at a time. So let's do [26:55] our text. We're editing the text here [26:58] and it switches to the way to start your [27:02] day. And here we can see if I zoom out a [27:05] little bit, we can now see the the [27:08] original and then the updated. And now [27:11] I'm going to now I'm going to select [27:13] here and then I'm going to send to make [27:16] a greener color. So we can see the [27:20] original and we can see the updated [27:22] version. And actually I want to just [27:24] make the greener color to the updated [27:26] version. So, I'm going to go back to the [27:28] comment tool and I'm going to say make a [27:31] greener color. And then I can apply [27:34] that. Okay. And here's the updated one. [27:36] The the way to start your day and with [27:39] the greener color. Another thing we can [27:41] do is mark. I'm going to select this [27:43] sun. And then I can actually just it it [27:46] puts over into our chat over here and I [27:48] can say make the sun larger. So [27:52] basically, you don't even need an image [27:53] editing software. You can just have AI [27:56] make all the edits that you want and you [27:58] just describe it with text. And here it [28:00] is with the bigger sun here. Another [28:02] feature I want to talk to you about is [28:05] meeting minutes. So I just click this [28:06] button right here, meeting minutes. And [28:09] Manis can process meeting recordings and [28:11] generate structured summaries with [28:14] speaker recognition. If you connect a [28:16] tool like TLDV for meeting recordings, [28:20] Manis can pull in the recording, [28:22] identify who said what, and produce [28:25] organized notes with action items. So, [28:27] basically, just click start and it will [28:30] start recording. Also, this would work [28:32] on the app, too, if you don't want to [28:34] have it open on your website. [28:37] This is useful if you're in a lot of [28:40] meetings and spend too much time writing [28:42] up notes afterwards. You can set up a [28:45] workflow where your meetings get [28:47] recorded and Manis automatically [28:50] produces summaries. [28:54] So you can see as I talk it's actually [28:59] it's showing a transcript right there. [29:02] So I'm just going to finish this and [29:04] we'll see what it comes up with. So now [29:07] it's just going to process the [29:09] transcript. Okay, we just got this [29:13] meeting summary. So, basically, uh, it [29:16] knows that there was one speaker, the [29:18] demonstrator, and yep, that's basically [29:21] what it was about. So, this is a really [29:24] cool feature for any content creation [29:26] task. Here are a few tips that will [29:28] improve your results. Be specific about [29:31] the visual style you want. Professional, [29:34] minimalist, bold, and colorful, [29:36] corporate. The more guidance you give it [29:38] on aesthetics, the better. Specify the [29:41] exact dimensions if you're creating [29:43] graphics for a particular platform. [29:46] Instagram has different dimensions than [29:48] LinkedIn, which has different dimensions [29:51] than a YouTube thumbnail. Give Manis the [29:54] actual text you want included. If you [29:56] want specific numbers, dates, names, or [29:58] taglines on your slides or graphics, [30:01] spell them out in your prompt. Don't [30:03] make the AI guess what your tagline [30:05] should be. [30:06] If you're creating a presentation, write [30:09] out a brief summary of each slide's [30:11] content. For example, slide one should [30:13] be the title with our company name and [30:15] logo. Slide two should cover the problem [30:17] our customers face. This level of [30:20] direction produces much better results [30:22] than make a pitch deck deck about our [30:25] company. [30:28] This is where things get really [30:30] interesting. Manis can build entire [30:32] websites, mobile apps, and desktop apps [30:34] from natural language descriptions. It's [30:38] not going to replace a professional [30:40] developer for complex applications, but [30:43] for landing pages, portfolio, small [30:46] business sites, and MVPs, it's [30:48] remarkably capable. So, let's build a [30:50] website. So, I'm just going to click the [30:52] button build website and then describe [30:54] what I want to build. I'm just going to [30:56] put in here build a landing page for a [30:57] dog walking service called Happy Pause [30:59] in Austin, Texas. It should have a hero [31:00] section with a headline called action [31:02] button, service section, three packages, [31:05] a pricing, about section, customer [31:07] testimonials, contact form, user [31:08] friendly warm color color scheme, make [31:10] it mobile responsive, and then I just [31:12] click submit. Manis is going to start [31:14] building. And just like all the other [31:17] demonstrations we saw, we'll be able to [31:19] watch it write code, create assets, and [31:21] assemble the page. Now, it's going to be [31:23] better with the Max 1.6 with 1.6 six [31:27] max, but I want to show you that it can [31:29] work with the the light plan also. It [31:32] just won't be quite as good. So, I'll do [31:34] try light anyways. So, Manis is going to [31:37] handle the HTML, CSS, JavaScript, and [31:39] it's going to build a functional [31:41] responsive website. And the task is [31:44] done. We can preview it right within [31:46] Manis. And as we're looking at this, [31:48] remember this is the the free plan. It [31:51] would be even better if we had the paid [31:53] plan, but you can see here's the website [31:56] and it actually generated all these [31:58] images for the website. If we scroll up [32:02] a little bit, we can see it generated [32:04] all the images that that it needed. And [32:07] so it's giving us different plans, the [32:09] about section, uh the customer [32:12] testimonial section, basically [32:15] everything that we asked for. So, we can [32:18] actually go to the different sections [32:20] and we can edit it just like how we [32:24] could edit the images and the [32:25] powerpoints. We can easily edit any [32:27] section of this website. So, I'm going [32:29] to go into this text content. I can add [32:32] new text here. Dogs are amazing. I can [32:35] even change the tech, the color of the [32:37] text. We can choose the the layout [32:40] margin. We can basically even we can use [32:43] AI to describe the change. And then we [32:46] can check to see if it is mobile [32:48] friendly. Yep, it is mobile friendly. [32:50] Let me just apply that change that I [32:52] created that I used. And we can see it's [32:55] going into the HTML to make the changes. [32:58] Okay. And so we can not only can we see [33:01] this preview, we can also go into the [33:04] code and see all the code, we can [33:06] download the code, we can go into the [33:10] dashboard. This is actually this is only [33:12] going to show after we publish the [33:14] website. And then some websites could [33:17] require a database and we can actually [33:20] see uh connect to the database on this [33:23] tab. And then we can also connect to [33:26] file storage and then we have settings [33:29] for our website including connect uh [33:32] publishing it, making a copy and then we [33:36] can even um connect different our custom [33:40] domains and then we can even set up a [33:43] payment integration. We can set up [33:45] Stripe through here and then we can have [33:48] secrets. We can even um integrate it [33:51] with GitHub. So I'm going to publish [33:54] this website and just click publish now. [33:57] And now it's already live. I can press [34:00] visit. And here is the website live for [34:04] anybody to visit. It's at manis.space. [34:07] But now that we've published it, I can [34:09] click custom domain and we can add any [34:13] custom domain or even buy a domain right [34:16] from this interface here. So basically, [34:20] Manis has built-in hosting and custom [34:23] domain support. Also, Manis handles SSL [34:27] certificates automatically, so your site [34:30] will have HTTPS from the start. And then [34:33] I'm going to talk a little bit more [34:35] about the SEO here. Manis has some solid [34:38] built-in tools. Now, you need the [34:41] upgraded planned to access these, but it [34:44] uses a dual delivery system where real [34:47] visitors get the full interactive page. [34:51] But search engine crawlers get a clean [34:53] static HTML version that's easy to [34:56] index. It automatically generates [34:58] robots.txt files and sitemaps. And [35:02] there's this in this SEO dashboard, it's [35:04] going to show a health score for the [35:06] site. And another thing about the [35:09] payment, I can click start payment [35:11] integration. And one really interesting [35:14] thing about Manis is that you can add [35:17] Stripe payments to your Manisbuilt [35:19] website without even having a Stripe [35:22] account. This is something that I have [35:24] not seen in any other AI website [35:26] builder, and I've used quite a few of [35:29] them. So, here's how it works. When you [35:31] ask Manis to add payment functionality, [35:33] it creates what's called a Stripe [35:36] claimable sandbox. This is a test [35:38] payment environment where everything [35:40] works but no real money moves. Manis [35:43] sets up the products, configures the web [35:46] hooks, and runs end toend tests to make [35:49] sure everything functions properly. So, [35:52] when you're ready to go live and accept [35:54] real payments, you claim the sandbox, [35:56] which converts it into a permanent [35:58] Stripe account. It's a really smooth [36:00] process. So right now it's upgrading [36:03] HappyBots to a full stack project with [36:05] database support with backend [36:07] capabilities. It's also enabling the [36:10] Stripe payment processing. So we'll come [36:12] back to it once it finishes. And it's [36:15] done. And I also accidentally clicked [36:17] one of the recommended um additional [36:19] prompts. So it's now also making the [36:22] process we use here into a re reusable [36:24] skill with skill creator. So it's [36:27] creating a reusable skill that documents [36:28] the Stripe integration process we just [36:30] performed. So while it's doing that, [36:33] let's check out the website. So I'm [36:36] going to publish the latest version of [36:39] the website. And now I'm going to visit [36:41] it. So now let's see how this works. I [36:44] can book a walk today. So now I'm going [36:47] to choose one of these plans here. And [36:49] we're having a problem, but that's okay. [36:51] It's very easy to fix with Manis. So I'm [36:53] back over at Manis. and it's and it did [36:54] finish creating this skill. Basically, [36:58] now it it's going to make it a lot [37:00] quicker to add Stripe payment [37:03] integration to another website. So, you [37:05] can create skills that will make things [37:07] that that will basically save on credits [37:11] uh when when you have to do the same [37:13] thing again. I'm just going to click add [37:14] my skills. And now I'm going to tell it [37:17] to fix the website. Okay, I'll just tell [37:19] about the error and I'll just wait for [37:20] it to fix the error. [37:23] Okay, I went back and forth a little bit [37:25] and fixed some Stripe errors. So, let's [37:27] test it out. I'll just go to the buy now [37:31] page. [37:33] I'll just get my test card information [37:36] because it got set up with the Stripe [37:38] Sandbox and I'll be able to use this [37:40] now. I'll just click the buy now link [37:44] and yep, it opens my Stripe integration. [37:47] So, I'll just go and fill this out with [37:50] the sample demo data just to make sure [37:54] this works. And then I'll click pay. [37:58] Okay. And now it's showing my bookings. [38:01] There's three here because I did some [38:03] tests before that. But and it's also [38:06] showing payment successful. So, your dog [38:10] walking booking has been confirmed. Will [38:12] confir will contact you to schedule the [38:15] walk. And this works and it's showing [38:19] the date, all the information, who made [38:22] the booking. So, so I was able to make a [38:27] full website with Stripe payments and [38:31] authentication [38:33] with just a little bit of time going [38:36] back and forth with Manis. Now, if we [38:39] scroll up in the chat here, we can see [38:41] that it gives some next steps to [38:43] consider, including claim your Stripe [38:46] sandbox. So, if we click this link, we [38:50] can actually log into Stripe and claim [38:52] it to our Stripe account and basically [38:56] have a live uh a working live Stripe [38:59] payments on our site. Okay, now let's [39:02] talk about mobile apps. Yes, Manis can [39:04] build mobile apps, too. You describe [39:06] what you want the app to do and Manis [39:07] generates it. So, I'll just type build a [39:09] simple habit tracker mobile app. Users [39:12] should be able to add habits, check them [39:14] off daily, and see a weekly streak [39:18] counter. Keep the design clean and [39:20] minimal. And then I can just select that [39:23] I want to develop apps. And then I click [39:26] submit here. And you can see it will [39:28] just start planning the app. And it will [39:31] give me a list of tasks that it's [39:34] working on. And then I can also click on [39:36] Manass's computer to see what it's doing [39:38] on its computer. So this is going to [39:41] take a little while, but you can see [39:43] that it starts by creating a plan. So [39:48] this whole time I can just watch what [39:50] it's doing, including creating the plan, [39:53] creating the code, and it's even going [39:56] to create images to use in the [39:58] application. But I'll just come back [40:01] later when it's actually completely [40:04] finished [40:07] and it's done with the application [40:11] and it gives us a description of [40:13] everything it's done. So now let's try [40:17] it out. Now we can try it out right in [40:19] our browser here. I'm just going to add [40:22] some task, a workout task [40:25] and say that I completed it and then [40:27] just trying adding another task about [40:29] drinking water. [40:32] One thing to know is just like it says [40:34] here, some features aren't available in [40:36] your browser. So, we can actually [40:38] preview this on on our phone. So, I'm [40:42] just going to take out my phone and scan [40:45] this QR code and then go through the [40:48] steps to get this installed to try it [40:51] out on my phone. Okay, now I'm checking [40:54] this out right on my phone. So, I'm [40:56] going to try adding a new habit. How [40:58] about practice instrument and add the [41:01] habit? And then one more. I'll choose [41:04] the icon and then just put clean [41:06] kitchen. And so you can already see that [41:09] I I can see a lot more of the [41:11] application on the mobile app than I [41:14] could on my phone. So I can mark these [41:16] complete and then I can also see the [41:18] streak page just one day so far. So this [41:22] is pretty cool. I can even click on the [41:25] stats to get additional stats about the [41:28] application. So this is not a bad app [41:31] that I just created just with a single [41:33] prompt. and the entire app uh we can [41:36] test out but we can also see the code. [41:38] So if I click on this code here we can [41:40] see the code and we can even download [41:42] the code if we want to just edit it [41:45] locally. And if I click publish up here [41:49] uh manage will build the app for Android [41:51] it generates an AAB file. That's the [41:54] standard Android app bundle format. For [41:56] iOS it can generate a build and target [41:59] and autoupload it to test flight for [42:01] testing. An important important note [42:03] here, Manis does not publish apps [42:05] directly to the App Store or Google Play [42:07] Store. It gets you to the testing. [42:10] You'll still need to go through the [42:12] actual submission and review process [42:14] yourself, but it handles the building [42:16] part, which is often the hardest part [42:18] for non-developers. So, when should you [42:20] use Manis for building versus a [42:22] dedicated tool like Web Flow, Shopify, [42:25] or WordPress? Well, here's my take. [42:27] Manis is excellent for quick prototypes [42:30] and MVPs. If you need a landing page up [42:33] to date, Manis can do it. If you need if [42:35] you're testing a business idea and need [42:37] a special site with payments, uh you can [42:39] use Manis. But for complex ongoing [42:41] websites with lots of content, like a [42:43] blog with hundreds of posts or an [42:45] e-commerce store with thousands of [42:47] products, you're better off with a [42:48] dedicated platform. Those tools are [42:51] built specifically for ongoing content [42:53] management. For mobile apps, MANS is [42:55] great for simple utility apps and [42:57] prototypes. If you need a complex app [42:59] with features like real-time messaging, [43:01] complex animations, or offline sync, [43:04] you'll need professional development. [43:06] Think of Manis as the fastest way to go [43:08] from idea to something real. It won't [43:11] always be the best tool for the long [43:12] term, but it's often the best tool to [43:14] get started. [43:18] One of the things that makes Manis [43:20] genuinely useful for day-to-day work is [43:22] its ability connect to other tools you [43:25] already use. These connections are [43:28] called connectors. To set up a [43:30] connector, go to the connectors section [43:32] in Manis. So, I'll just click the [43:34] settings in the corner here and then go [43:36] to connectors and then add connector. [43:40] And you'll see a list of available [43:42] services. [43:43] And all of these are services we can [43:46] connect and we can go through an [43:47] authentication process. So, for example, [43:50] I could connect uh Gmail to a to access, [43:55] search, and organize and reply to [43:58] emails. And then I'll just click [44:00] authorize and connect. Now, it's going [44:02] to I can click the try it out button. [44:04] It's going to give me a prompt to try [44:06] out. Help me test the Gmail connector [44:08] and show me how to use it, its features. [44:10] So, submit that and it's going to think [44:13] for a while. [44:15] Now I can click the dropdown for [44:17] searching Gmail and these are actual [44:20] emails from my account. So it was able [44:23] to actually access my email. Okay, I'll [44:26] go back to connectors again and I'm just [44:28] going to disconnect my email because I [44:31] think I don't really want it connected [44:33] right now. But let's look at some of the [44:35] other connectors. Google Workspace is [44:37] probably the most practical connector [44:39] for most people. Uh once it's connected, [44:42] man can work directly with your Google [44:44] Docs, Sheets, and Slides. It can read [44:46] existing documents, edit them, um create [44:49] new ones, and organize files in your [44:51] Google Drive. Now, Instagram and Meta [44:53] Ads Manager are available if you're [44:55] managing social media or running ads. [44:57] With Instagram connected, you manage can [44:59] help you ideulate content, generate [45:01] images, write captions, and even publish [45:03] posts. With Meta Ads Manager, it can [45:06] analyze your ad performance and give you [45:08] insights. For other connectors, there's [45:10] Google Drive for file access, Notion for [45:12] birectional data flow through the MCP [45:14] port protocol. We have Stripe for [45:17] payment integration like we covered [45:19] earlier, uh, Telegram, just a bunch of [45:22] different ones. I'm actually going to [45:24] connect to Google Drive and Slack. So, [45:28] it's going to give it's going to give [45:29] kind of instructions when you're going [45:30] through it how to connect to Slack. And [45:32] I'm going to show you how a multi-tool [45:35] workflow can bring several connectors [45:38] together. So, it's really practical. [45:41] I'll type search Google Drive for the [45:43] file called Q1 customer feedback. Read [45:44] through the feedback. Categorize each [45:46] piece into themes. Count how many [45:48] mentions of the seam. Create a summary [45:50] spreadsheet with the categories. Uh some [45:52] other information. And then also post a [45:54] summary to the product feedback channel [45:56] in Slack. And I'll submit. And while [45:59] it's working on this, let me show you [46:01] this is what the Google doc looks like [46:04] that's going to be accessing. So it's a [46:05] lot of information that's going to have [46:08] to evaluate. So it's thinking and it's [46:12] it's just going going through a few [46:13] different tasks. And you can see it's [46:15] actually writing code to uh help with [46:18] the tasks. That's something that not all [46:20] AI tools can do. And here it is. It [46:23] saved this spreadsheet in in my Google [46:26] Sheets. So here's the summary that it [46:29] created. And now you can see it's trying [46:31] to post to Slack. It sees that the [46:33] product feedback channel does not exist. [46:36] So it's basically just going to figure [46:38] out how to solve the problem itself. [46:40] It's going to basically post a summary [46:42] in a different channel. I don't even [46:43] have to tell it. It kind of figures out [46:45] what's the best way to accomplish what [46:47] the user said. Even though I'm running [46:49] into some problems. Okay, it's giving me [46:52] the results here in Manis. It gives a [46:54] link to the spreadsheet. Now, let's [46:57] check out over on Slack. Yep. See, this [47:00] is what it gave me right in Slack. It [47:01] gave me a full summary of my file. And [47:05] then it gives a link right to the Google [47:07] doc. So, this is a pretty nice [47:09] cross-platform workflow. Each individual [47:12] step isn't that hard to do manually, but [47:15] stringing them all together takes time [47:16] and effort. And Manis handles the full [47:19] chain. And there's plenty there's plenty [47:22] of ways you could use this. Like another [47:23] example, let's say you run a weekly [47:25] meeting and want to automate notes dist [47:27] distributed to your team. You could say [47:29] something like after each Monday [47:30] standup, put the recording uh pull the [47:33] recording from TLDV, generate meeting [47:35] minutes with action items organized by [47:37] team members. Uh save the notes to [47:39] Google Docs in the meeting notes folder [47:40] and post a summary with action items in [47:43] Slack. So this is the kind of [47:44] reoccurring workflow that can save you [47:46] 30 minutes every week and over a year [47:48] that can add up for connecting services [47:51] that aren't in the built-in connector [47:53] list. Manis supports custom API [47:56] connections and model contacts protocol [47:59] or MCP integrations. MCP is an open [48:02] standard that lets AI agents interact [48:05] with external tools in a standardized [48:07] way. If a service you use supports MCP, [48:10] you can connect to it to connect it to [48:12] manis even if there isn't a dedicated [48:14] connector. So basically in the connector [48:17] section, we just click custom MCP and [48:21] then uh add custom MCP and import by [48:24] JSON or you can do the direct [48:26] configuration and then they'll you'll [48:28] just have to put in all the information [48:30] about it. Setting up custom connections [48:32] is more advanced and usually require [48:34] some technical knowledge, but it means [48:36] Manis's integration capabilities extend [48:38] well beyond the built-in list. A few [48:40] practical tips for working with [48:41] connectors. Uh only only connect [48:44] services you actually plan to use with [48:45] Manis because each connection gives [48:47] Manis access to that platform. So, you [48:49] really want to be intentional about it. [48:51] When you're giving Manis a task that [48:53] uses a connector, mention the specific [48:56] service in your prompt like search [48:58] Google Drive for is better than find my [49:01] file about because it tells Mass exactly [49:03] which connector to use. You should also [49:06] test your connections with a small task [49:08] before relying on them for anything [49:10] important. Make sure mass can actually [49:12] read and write to the service the way [49:14] you expect. [49:19] So far, we've been running prompts from [49:21] scratch, but Manis also has has a [49:23] library of over 100 pre-built workflow [49:26] templates called the playbook. These are [49:28] ready to use prompts designed for common [49:30] tasks. So, let me show you how you [49:33] access the playbook. Basically, right [49:35] from the main mana screen, we can go to [49:37] more, then playbook, and it'll open into [49:40] a new tab here. You can browse this [49:43] playment and the templates are organized [49:45] into different categories. So you can [49:47] see that we have a market research tool, [49:50] we have AI video generator, we have um [49:53] AI text generation. You can see all [49:55] these and there's a lot of different [49:56] categories for different templates that [49:59] you can use. So let me show you how to [50:02] use one. I'm going to try this employee [50:04] directory generator. So, I'll just fill [50:06] in the blanks here with the company [50:08] name, uh, employee count, um, [50:12] departments, and then I'll just click [50:14] create directory, and they'll wait and [50:16] it's going to just start doing some [50:18] research to create the directory all on [50:20] on its own. I didn't have to give it any [50:22] information really besides the company. [50:24] So, it's going to finish planning and it [50:27] finished. It actually created an entire [50:29] website for the company directory. So, [50:32] we can go through it here. Now, this is [50:34] pretty cool. It uses a coding theme and [50:37] it matches a lot of the Free Code Camp [50:40] colors and uh we have all these cards [50:44] for different people that work at Free [50:46] Code Camp, including me. Here's mine. [50:49] And it has a correct bio. So, you can [50:52] see that uh the it created this entire [50:57] employee directory with themes. It even [51:00] uh created this image of the different [51:02] countries. I think these are actually [51:04] accurate to the the countries that [51:07] people actually work work at at Free [51:09] Code Cam. And that's an image that I I [51:11] saw Manis actually create. Okay, the [51:14] next one I want to try is the AI [51:16] interior design. So, I'm going to upload [51:19] a real image of my room and then just [51:23] stay my style pre preferences modern and [51:26] start designing. And you can see it has [51:28] these words. Uh, this is my living room. [51:31] Please help me redesign the layout. [51:32] Please search on furniture website like [51:34] IKEA and just has some more information. [51:36] And I can actually change this prompt if [51:38] I want. But I'm just going to click [51:40] start designing. And wow, this is what [51:43] it came up with. This is actually pretty [51:45] nice. I I like this. I could I could [51:48] seriously imagine using this product to [51:50] help design my my rooms. And if I click [51:54] into this document here, we can see more [51:56] information. It has real links. So, if I [51:57] click on the link, it opens the IKEA [51:59] page where I could actually buy the [52:01] couch right from the picture. So, yeah, [52:04] this is very helpful. So, yeah, you [52:06] know, there's just a bunch of other [52:07] templates and and playbook items that [52:10] you can try out. Now, the key benefit of [52:12] the playbook is that it saves you the [52:14] effort of figuring out how to prompt [52:16] Manis for common tasks. Someone has [52:18] already figured out the right way to to [52:19] phrase the request for best results. You [52:22] just plug in your details. So, if you're [52:24] new to MAS, I'd actually recommend [52:26] browsing the playbook before trying to [52:27] write your own prompts from scratch. It [52:29] it can give you a good sense of what [52:31] Mass can do and how to structure your [52:33] requests. Many of the templates also [52:35] include examples of the bracket pattern [52:37] we talked about earlier. [52:42] As you use Manis more, you'll find [52:44] yourself wanting to organize your work. [52:46] That's where projects come in. A project [52:49] in Manis is basically a workspace that [52:51] groups related tasks together. If you're [52:54] working on a marketing campaign, for [52:56] example, you might have separate tasks [52:59] for competitor research, content [53:01] creation, social media scheduling, and [53:03] analytics. A project lets you keep all [53:06] that in one place. Okay. To create a new [53:10] project, I'll just click this new [53:11] project button. And then we'll just [53:14] enter the name and the instructions. All [53:17] tasks in this project are for Acme Corp, [53:19] a B2B SAS company in the healthcare [53:21] space and then just has some information [53:24] here. Now when you set up a project, you [53:26] can write a set of instructions that [53:28] automatically apply to every task within [53:30] the project. Now every task you create [53:32] within the project already knows this [53:34] context and you don't have to repeat it [53:36] every time. You just focus on the task [53:38] at hand. And you can see now we have [53:41] this Acme Corp project and every task [53:45] that we create within this project will [53:48] have access to the instructions, the [53:50] files, the skills, the scheduled tasks [53:53] and the task will know about each other. [53:55] So you can kind of collect all your [53:57] information in one place for a project. [54:00] It's basically like a know a knowledge [54:02] base here. You can upload files that [54:04] should be available to all tasks in the [54:05] project like brand guidelines, project [54:08] specifications, customer personas, uh or [54:10] previous reports. Any task within the [54:12] project can reference these files. Now, [54:15] skills here uh a skill is a reusable [54:18] piece of expertise or workflow that you [54:21] can create once and use across multiple [54:23] tasks. Think of them as saved [54:26] procedures. And then once you've created [54:28] a skill, you can apply to any new task [54:30] instead of reexplaining your entire [54:32] audit process each time. So I'm going to [54:34] go through these skills and I'm going to [54:36] choose uh video generator. And then I'm [54:39] going to say I'm just going to say use a [54:41] video generator skill to create a short [54:43] ad for my SAS. And it already has [54:46] information about the SAS from the [54:48] instructions and it knows about the [54:50] skill. So, let's just submit that and [54:52] see what happens. [54:55] Okay, it finished creating the video. [54:59] Let's check it out. [55:00] >> Managing a hospital means balancing [55:02] countless details. [55:04] Acme Corp streamlines your operations. [55:08] Sink and focus on what truly matters. [55:10] Improving patient care. Acme Corp, [55:13] empowering healthcare. [music] [55:17] >> Okay, not bad. based on giving very [55:20] little information uh for what I wanted. [55:23] I'm actually pretty happy with that. Um [55:26] and you can see it created all these [55:27] assets along the way. See, it created [55:30] the character so I could use a [55:31] consistent character throughout the ad. [55:34] Um so I've actually used other AIs that [55:37] are specifically for creating video ads [55:40] and this is actually better. Okay, let's [55:44] move on and let's cover team [55:46] collaboration. Manis has team plans with [55:50] features designed for organizations. So [55:53] we can go to get team plan and we could [55:57] get to get a team plan. On a team plan [56:00] you get shared projects and skills. [56:02] Everyone on the team can access the same [56:04] workspaces and workflows. There's single [56:08] sign on support for enterprise [56:09] authentication or author authentic [56:12] authentication and you get shared credit [56:15] pool so the team doesn't have to manage [56:17] individual credit balances. Also, [56:20] there's a collab feature. So, I'm back [56:22] on where I created the video. I can [56:25] click the button up here that says [56:27] collaborate and add different email [56:30] addresses. Collaborators can message in [56:33] this task using your credits. So [56:37] basically this allows up to 50 people to [56:39] part participate in a shared Manis [56:41] session. The person who creates the [56:43] session pays the credits. This is useful [56:46] for meetings or workshops where everyone [56:48] wants to interact with Manis together. [56:51] One thing to be aware of with shared [56:53] sessions is that collaborators in a [56:54] shared sandbox can see files and data [56:57] generated during the session. So just be [56:59] thoughtful about what information gets [57:01] surfaced if you're working with [57:03] sensitive data. For teams adopting [57:05] Manis, here are some suggestions. Start [57:07] by identifying the repetitive tasks your [57:10] team does every week. Things like [57:12] compiling reports, processing data, [57:15] researching topics, cing presentations. [57:18] These are the best candidates for Manis [57:20] workflows. Create projects for each [57:22] major initiative or or client and write [57:25] clear master instructions that capture [57:27] the important context. Build skills for [57:30] your most common workflows and make them [57:32] team level so everyone benefits. [57:34] Designate someone on the team to be the [57:36] manis champion who helps others learn [57:38] the tool and maintains the shared skills [57:40] library. Also, don't try to automate [57:43] everything at once. Pick two or three [57:44] workflows, get them working well, and [57:46] expand from there. [57:51] If you're a developer or you want to [57:53] automate Manis programmatically, there's [57:56] a full REST API available at [57:58] open.mmanis.ai. [58:00] The API lets you do everything the web [58:02] interface does but through code. You can [58:04] create tasks, check their status, send [58:07] follow-up messages, manage files, set up [58:09] web hooks, and more. This is how you'd [58:12] integrate Manis into your own [58:13] applications or build automated [58:16] pipelines. Okay, let's walk through the [58:18] basics. First, authentication. You need [58:21] an API key. So, we're going to go to the [58:24] settings in the bottom left corner and [58:26] then integrations and then build with [58:29] the Manis API. Now, we're just going to [58:32] click create new API key. And I can call [58:35] anything I want. I'll just call it [58:37] secret. And then you just need to copy [58:39] the API key. To authenticate your API [58:42] calls, include the key in the header. [58:44] The header name is X-manus- API- key, [58:48] not the typical bear token format you [58:51] see with other APIs. Keep your API key [58:53] secure. Treat it like a password. Store [58:55] it in an environment variable, not your [58:57] code. Now that we have the API key, [58:59] let's see how it works. I'm going to be [59:02] doing the API using the API with curl in [59:05] the terminal, but you could easily use [59:07] this with any programming language. The [59:10] API allows you to programmatically [59:13] create and manage AI agents, enabling [59:16] you to build automations, orchestrate [59:18] multi-step workflows, and integrate [59:21] Manis directly into your application. [59:24] Also, I'll be showing my actual API key [59:27] in this video, but before I post this [59:30] video, I'm going to deactivate the API [59:32] key. So, don't worry about me exposing [59:35] the key. So here's how you authenticate [59:38] a request using curl. First of all, all [59:42] requests are going to this a [59:43] api.mmanis.aiv2. [59:47] Here we're doing a just a create request [59:49] and we're just sending hello manis. We [59:51] can see the API key right in here. So I [59:53] will submit that. And it says okay, [59:57] true. And it's giving me a task URL. So [60:00] I can just copy that URL. [60:04] And we can see we created a new task. It [60:07] just says hello manis. That's what we [60:09] sent. And it's going to give us the [60:11] response that we sent. So let's talk [60:15] about the task life cycle. The core of [60:18] the Manis API revolves around tasks. [60:21] When you create a task, the agent runs [60:24] asynchronously. You can track its [60:26] progress by pulling from messages or by [60:29] setting up web hooks. So here's another [60:32] curl command. And I'm going to show you [60:35] basically how to start a new task. We're [60:38] using task.create. [60:40] This is actually what we did in the last [60:41] example also. And you can pass plain [60:44] text or an array of content parts [60:47] including files and connectors. In this [60:50] case, we're just putting the text and [60:53] we're saying in the message research the [60:55] top five AI trends for 2024. And we're [60:58] passing the title. So just like before, [61:01] it's going to say okay, true. we get the [61:03] task URL. So we can go to the I'm going [61:06] to copy the URL and we can see the URL [61:08] and we can see that it's currently [61:10] working. So we want to pull for [61:13] progress. Basically once a task is [61:15] created we can create it. We can [61:17] retrieve its event history using the [61:19] task.list messages endpoint. So I'm [61:22] going to get this task ID right here. [61:24] This is how we can re reference the task [61:27] and it also happens to be right in the [61:29] URL. So, let me paste in another curl [61:32] command. We have our API key like [61:34] before, but now we have this task ID. [61:39] So, I'm just going to submit that. And [61:42] it's going to give us a lot of [61:43] information here. One thing we want to [61:45] look at is status update. So, the agent [61:48] status field indicates the current [61:50] state. In this case, running. That means [61:53] the task is still running. If we go back [61:56] to the browser, we can see that it's [61:58] still working. it it's still trying to [62:01] read some information to run the task. [62:03] The status could also say stopped if [62:06] it's completed, waiting if the agent [62:08] needs user confirmation or input or [62:11] error if the task has failed. And you [62:14] can see it's also giving the message [62:16] from the assistant that that it's [62:19] starting is what it's going to do in [62:21] starting the research. That's also what [62:22] we see right in the browser here. So [62:26] when the agent status is waiting, it [62:29] requires input to proceed. And there's a [62:32] few different types of waiting states. [62:35] One is message ask user. The agent is [62:37] asking a question. In that case, you can [62:41] send a message just like we did before. [62:43] In that case, we would send do task. [62:47] Send message. And then you would have to [62:49] put in your API key. I didn't put it in [62:51] this time. And then also your task ID. [62:54] Right now, it's not asking for anything, [62:56] so I can't show you an example. But [62:58] there's other waiting types like [63:00] confirming an email or or uh basically [63:04] sending or confirming a deployment. So, [63:07] you must use the task.confirmaction [63:10] endpoint. So, that's going to look like [63:13] this. Task.confirm [63:16] and then we would put the API key, the [63:17] task ID, and then the event ID, and then [63:21] say input accept true. If we're going to [63:24] confirm that we want to do something and [63:27] you can also work with files, you can [63:29] upload files be processed by the agent [63:32] using the upload file endpoint. So we [63:36] have file.upload [63:38] and then we have our API key and then we [63:40] just pass in the file name file.csv. [63:43] Basically we're in the middle of this [63:45] two two-step process. Running this is [63:48] going to basically [63:51] create a file record to get an upload [63:54] URL. So here is our upload URL here. And [63:58] now we need to upload the actual bytes. [64:02] So I'm going to put another curl command [64:04] and it's going to put to the upload URL [64:08] and then binary data the file.csv. [64:10] That's the file. Oh, I have the file [64:13] name wrong. So I'm going to try that one [64:14] more time. Okay. Hey, and I had an error [64:16] before cuz I had these double quotes [64:18] around the URL, but I needed to switch [64:20] it to single quotes. So, now that our [64:22] file is uploaded, and once uploaded, we [64:25] can include the file ID in the [64:26] message.content array when calling the [64:29] task.create or task.s send message. Now, [64:33] I'm going to run my task.list messages [64:36] again for one of our earlier tasks, the [64:40] one that Manis was still working on. And [64:43] now we have the results. This was the [64:46] top five AI trends from 2024. [64:49] Now, it's kind of hard to read. Uh you [64:51] can see that it is formatted [64:54] with markdown and we have new lines, but [64:56] in the terminal always it just kind of [64:58] looks like one big block of text. So, [65:01] this is what it's going to look like if [65:02] we go over to the the the Manis website. [65:06] So basically this is was the result and [65:09] we got the markdown of this result right [65:11] here and we could easily if we were [65:13] using this doing this in code we could [65:15] easily format this format the markdown [65:18] so it's easier to read. Now this use [65:20] projects projects allow you to group [65:22] related tasks and apply shared [65:24] instructions or system prompts [65:26] automatically. This is good for [65:28] enforcing consistent behavior such as [65:30] specific persona or output format. So [65:32] this is how you create a project. We [65:34] have a project.create. Then we have the [65:36] name of the project and then we have the [65:39] instructions and I could submit that. [65:43] And now it's going to give us an ID. [65:46] This this is the project ID which it you [65:49] can then pass as project ID when [65:51] creating new tasks and the instructions [65:54] will be applied automatically. [65:57] You can also use web hooks for real time [66:00] notifications instead of pulling [66:02] tasks.list lists messages which consumes [66:05] rate limits. It's highly recommended to [66:08] use web hooks for production [66:09] applications. Web hooks send HTTP post [66:13] requests to your server when tasks are [66:15] created or when their state changes like [66:19] completing or waiting for input. You can [66:21] configure web hook hooks in the Manis [66:24] API integration settings. Your endpoint [66:27] must be able to receive JSON payloads [66:30] and respond with a 200 okay status [66:32] within 10 seconds. So this is an example [66:35] of a simple web hook handler in Node.js [66:38] and Express. And if we go into our [66:41] settings, our integrations and here is [66:45] where you can create a new web hook. So [66:49] you just basically enter your URL here [66:52] and then you can use your web hook in [66:53] your application. The Manis API enforces [66:56] per user rate limits across all [66:59] endpoints. For example, task.create is [67:02] limited to 10 requests per minute while [67:05] task.list messages allows 100 requests [67:08] per minute. When designing your [67:10] integration, you should uh make sure to [67:13] use some best practices like use web [67:15] hooks u handle 429 errors. If you [67:18] receive an HTTP 429 rate limited [67:21] response, you should need to implement [67:23] exponential backoff with jitter rather [67:26] than retrying immediately. Also use [67:28] pagionation. You can use cursor [67:30] pagenation uh cursor and limit [67:32] parameters on list endpoints to manage [67:34] large data sets efficiently. [67:40] Before we wrap up, I want to be honest [67:42] about where manis works well and where [67:44] it falls short. Understanding the [67:46] limitations is just as important as [67:49] knowing the capabilities. [67:51] Manis is slow compared to chat bots. [67:54] When it's browsing the web, creating [67:55] files, and running code, it takes real [67:58] time. A research task might take 5 to 10 [68:00] minutes. A complex website build could [68:03] take longer. This is the trade-off for [68:05] getting real actionable output instead [68:07] of just text. Manis can get stuck. [68:10] Sometimes the browser automation hits a [68:12] wall. It might loop on a page, fail to [68:15] navigate a tricky website, or struggle [68:17] with unusual interface elements. If you [68:20] notice a task spinning without making [68:22] progress, you can stop it and try [68:24] rephrasing your request. The quality [68:27] varies. Some tasks come out great the [68:30] first try, others need refinement. Treat [68:33] everything manage produces as a first [68:34] draft and plan to review it. This is [68:37] especially true for anything involving [68:38] numbers, dates, or claims of fact. The [68:41] AI is confident even when it's wrong, so [68:44] always verify the important stuff. Slide [68:47] presentations look great, but have [68:48] trade-offs. Nano Banana Pro slides are [68:51] beautiful, but harder to edit. HTML [68:54] slides are editable, but less visually [68:56] impressive. Pick based on your needs. [68:59] Apps built with Manis are prototypes, [69:01] not productionready software. They're [69:04] perfect for testing ideas, but if you [69:06] want a want to build a real product, [69:08] you'll eventually need proper [69:10] development. The free tier is limited. [69:13] The daily credit allowance goes fast, [69:15] especially with researchheavy tasks. If [69:18] you're serious about using Manis [69:20] regularly, you'll probably need a paid [69:22] plan. Manis is an AI agent that executes [69:26] tasks, not just answers questions. It [69:29] runs tasks in isolated cloud computers [69:31] called sandboxes. You can watch it work [69:33] in real time. Browser operator lets [69:36] Manis access login required sites [69:38] through your local browser. For content, [69:41] manage manis creates presentations in [69:43] two formats. Generates and edits images [69:46] through design view and handles meeting [69:49] minutes. For building, it can create [69:51] websites with hosting, custom domains, [69:53] SEO, and Stripe payments. It can build [69:56] mobile apps for testing purposes. [69:58] Connectors link Manis to Google [70:00] Workspace, Slack, Instagram, Meta Ads, [70:03] and other services. Multi-tool workflows [70:07] chain actions across platforms. The [70:09] playbook has 100 plus pre-built [70:11] templates for common tasks. Projects [70:14] organize your work. Skills create [70:16] reusable workflows. Team plans enable [70:19] collaboration. The API lets developers [70:21] integrate Manis into their own [70:23] applications and build automated [70:25] pipelines. If you want to keep learning, [70:27] there's Manis Academy, which has [70:30] structured courses for different skill [70:32] levels and industries. It's free to join [70:35] and includes beginner through advanced [70:38] content. The best way to actually get [70:40] good at Manis is to use it. Pick a real [70:43] task from your work this week, maybe [70:45] something that would normally take you [70:46] an hour or more, and give it to Mannis. [70:49] See how it does. Refine your prompt [70:52] based on the results. Try again. You'll [70:54] develop an intuition for how to work [70:56] with it pretty quickly. That's [70:58] everything. Thanks for watching and best [71:01] of luck applying Manis in your software [71:03] development workflows.