YouTube Deep Summary

When Manis AI launched, one thing we began to realize was that these interconnected AI agents can now honestly automate almost all the tasks we perform on our computers. Human computer interaction felt like it was disappearing. And in some cases, that's absolutely true. The use cases they've shown are impressive. Humans don't want to do something, so just ask an AI agent to handle it and it gets done. These agents have access to all sorts of tools like a browser and local file access. Think about it. What do you really do on your computer? You browse, process information, write it down somewhere or apply it somewhere else. That's exactly what these AI agents are doing. People are asking them to perform real tasks like building tax policy visualization tools or writing code. You can also ask them to do your taxes if the right data is given to them. The basic plan is $19 per month. It gives you 1,900 total credits a month and 300 refresh credits per day. Credit usage adds up fast. Making a simple chart can cost up to 200 credits, and that's most of your daily quota gone in one go. Developing a web app costs 900 credits, something you can't even do on the basic plan. So, if all you want is an AI agent to perform tasks on your computer, Manis becomes pretty costly very quickly. That's why I came across this open-source alternative called Agentic Seek. It basically does what Manis does, but it's completely local. If you've got a powerful enough computer, you can run it entirely on your machine without paying anything. No subscriptions, no credit limits. This is the power of open source and it's super exciting. In this video, I'll show you how Aentic Seek works. I'll walk you through real examples of the agent in action. And if you want to try it yourself, I'll show you exactly how to set it up. Don't worry if you don't have a powerful computer, you can still use Agentic Seek with an API connection, which is still far more affordable than Manis. First, let me show you a demo they've shared to give you a glimpse of how the agent works. Okay, so here you can clearly see they ask their own agent to search for the Agentic Seek project they're currently working on. The agent first needs to figure out what skills are required for the project. Then they provide it with a candidate zip file and the agents task is to find the candidates that best match the project requirements. The workflow is broken down across multiple specialized agents. One of them is the planner agent. First, it decides to go to the GitHub repository to identify the Agentic Seek project and determine the required skills. It reads the readme and the project description to extract that data. For some reason, it also fetches a comparison between Agentic Seek and Manis AI. Not entirely sure why it does that, but it's part of its planning flow. After gathering the necessary information, the planner agent proceeds to extract the contents of the CV candidate zip file. Once extracted, it navigates into the folder and starts reading the candidate files one by one. You can see it going through different profiles. After processing the files, it compares the candidates skills against those required by the Agentic Seek project. It eventually concludes that the best matches are Aisha Khan, Longch and a few others as well. It even ranks them based on fit. Normally this would require a human to search for the project, gather the skill requirements and feed everything into a tool like chat GPT, but this agent automates the entire process from start to finish with no manual input and no prompts. And the best part, you don't need to pay for anything like Manis AI. If you have a powerful enough computer, it all runs locally, fully private, and completely free. That's what makes it so incredible. I've even got this set up here myself. And this is the prompt I gave it. It needs to search online for popular sci-fi movies from 2024 and pick three that I should watch tonight. It did exactly that. The planner agent came online, broke my request into smaller tasks. And since I asked it to save the results in a movie night text file, it's doing that as well. Over here, I was seeing the browser view showing how it was browsing and searching through the web. And this is what it did. If we come back, it actually saved the file where the agent folder was placed in my developer folder and automatically created the movies.txt file there. These are the movies it gave me. So, it's pretty awesome and works really well. I also did a timing analysis. The request we just made to search for movies and save them in a text file. I recorded the time and it took 4 minutes to complete the entire task. I felt three sources weren't enough, so I asked it to search 10 sources instead. It did that, searched all 10, and took about 8 minutes before giving me a report. Overall, it's a pretty flexible agent. The timing, in my opinion, is really good, considering it's going out to all these sites and gathering the data I asked for. If you're enjoying the video, I'd really appreciate it if you could subscribe to the channel. We're also testing out memberships to support the channel. We've only launched the first tier so far. It offers priority comment replies for now, but subscribing would really help us see how many of you are interested and want to support what we're doing. If you found this worthwhile and want to install it, here's what you're going to do. Come on to the main website. I'll have the link in the description below. From here, go to the GitHub page. There you'll find all the installation commands I'm about to show you. Most of them are already there. You just have to copy and paste them. Follow along with what I do and you'll have the agent installed and running. First, clone the repository. Just copy the first command and paste it down here. This clones the repository, navigates into the directory, and renames the example env file to your actual env file. Next, you're going to create a virtual environment for Python. Inside that folder, paste this command to set up your Python environment. The best thing about this agent is that it supports installation on Windows, Mac OS, and Linux. So, you're not limited by your operating system. These are the two install scripts that will set up all the dependencies. Choose the one that matches your OS. For me, it's Mac OS, so I'm going to paste this and it'll install all my dependencies. You might get an error when you run this install script just like I did. The error will ask you to install Python version 3.10. You'll need to install that specific version for everything to work. The method differs depending on your operating system, but it shouldn't be too difficult. For Mac OS, you install it using this command. Once that's done, you should be good to go. After Python 3.10 10 is installed and you've run the installation script, everything should be set up and you'll be ready to run the agent. Let me guide you through the configuration before running the agent because that's quite important depending on how you plan to use it. Next, go into your terminal and open the directory in any code editor you like. I'll open it in cursor. Inside the codebase, you'll find a file called config.ini. This file contains the configuration settings that the agent uses while running. You'll need to change a few things depending on how you plan to use it. You can either run the agent locally or use it with an external API. If you want to run an LLM locally, head over to the GitHub repository. It has all the details listed. First, it shows how to set things up for a locally running LLM. To get good performance with this agent, you need large models running locally. Without those, it's really not going to be useful. Running 14B models from Olama, Deepseek, or Quen won't help much here because the performance won't be great. You need at least a 32B model for it to work well. that requires an RTX 4090. Even the 14B models I mentioned need at least an RTX 3060, which is cheaper than the 4090, but still pretty expensive. If setting this up locally isn't an option for you, then move on to the API setup. Right now, it supports APIs from OpenAI, DeepS, Hugging Face, Together AI, and Google. These are the providers it currently works with. I hoped Claude would work with it. Support for Claude isn't available yet. Now go ahead and change the values in the config. First set is local to false. It's set to true by default. Then define the provider name. Choose any provider from the list. I chose OpenAI, so I set that as mine. Next, define the provider model. I recommend using GPT40 instead of GPT4 mini. I tested both and GPT40 performed a lot better. If you have access to other providers, the performance may be even better. I didn't have credits for Deepseek or the others, so I just used GPT40. It worked well and stronger models would only improve it further. These are the main settings you'll need to update. Then open your ENV file. In that file, paste your API key, whether it's from OpenAI, DeepS, or whichever provider you're using. Once that's done, your configuration will be ready for basic use. There are also some optional settings. One of them is for a headless browser. This means the browser window won't actually open while the agent runs. It'll still do everything but quietly in the background. Leave this set to true. You can also enable speak and listen modes. These let the agent talk back and listen to your voice. You'll be able to have a real conversation. Just turn both of those options to true and it'll start working. Before you start up the agent, there are a few things you need to know. You'll need to have Docker up and running for the agent to start because it's containerized. It fetches containers and sets them up automatically. First, start Docker. Once it's running, you can start the services. Next, come back into the GitHub repository. And here, you'll run this command. It's different for Mac OS and Windows. The Mac OS command also works on Linux. So, just run that if you're using Linux. If you don't have your Python environment activated and you've opened a new terminal, make sure to run the environment activation command first. This will start a few backend services as well as the front end. The structure they've built is a bit confusing to me because usually backend and front end are set up separately. Here the setup is different. After running that, you're going to start the back end which is handled by api. py. Let me show you. This is my back end. These are the services that are running. They're just Docker containers. I use this command and now they're all running. Another thing to keep in mind, you need to run both the front end and back end in separate terminal windows. So, this is running here and the Python API command is running in another terminal. Something else they got wrong is that it won't start up if you just use Python or Python 3. You have to explicitly write Python 3.10 API. py for it to work. Paste that in and it'll start the back end. Then go to this address, localhost 3000, and you'll find your agent there, ready to use. That brings us to the end of this video. If you'd like to support the channel and help us keep making tutorials like this, you can do so by using the super thanks button below. As always, thank you for watching and I'll see you in the next one.