00:00I have showcased multiple deep research agents on this channel from Abacus AI's deep agent or Manis AI all the way to Gen Spark

00:08All of these are great deep research agents, but they're not open source or local

00:12But I have a solution for you

00:14I have found a new deep research agent that is fully local, open source, and quite powerful all while being super easy to install

00:21Allow me to introduce Gearflow

00:23Gearflow stands for deep exploration and efficient research flow

00:29It's a community-driven deep research framework that builds upon the incredible work of the open- source community

00:34It combines large language models with specialized tools for tasks like web search, crawling, and Python execution

00:41And it could also be powered by MCPs

00:44And essentially, it's trying to mimic the same sort of capability that you would see with the deep research agents of Manis or Gen Spark

00:50Just take a look at it live in action

00:53It was requested to write a brief on the top one trending repos on GitHub and you can see right away it is able to take on this action pretty quickly

01:02It is able to compile the resources and you can see the top trending repo was Cortex AI Suna AI which is another deep research agent that is open source but in this case it is pretty hard to install

01:15In this particular example, it was able to research on this repository, provide the key points, the overview, a detailed analysis purpose, and then also provides images showcasing the actual repository, and then it also provides key citations

01:33Before we get started, I just want to mention that you should definitely go ahead and subscribe to the World of AI newsletter

01:38I'm constantly posting different newsletters on a weekly basis

01:43So, this is where you can easily get up-to-date knowledge about what is happening in the AI space

01:47So, definitely go ahead and subscribe as this is completely for free

01:50This is a multi- aent architecture where it first starts off with the coordinator

01:55It is something that's responsible for engaging with the user prompt and then it sends in that prompt to the planner which is able to use different AI agents to update the plan with the human feedback which is then sent back and forth in a cycle

02:10And once it has finished working on that, it uses the research team to work on developing and executing that plan

02:17Uses the research agent as well as the coder if there is any sort of coding needed

02:21And then it sends back all of this and the planner then sends it to the reporter and then the reporter then sends it back to you

02:30Now the main focus is is that you get the ability to use open-source models with Olama

02:34So it's good for models that have deep research integrated like the new deep uh seek R1 or the Quen model

02:42You can also use open AI compatible APIs, multi-ter LM systems for different task complexities

02:47You have tools in MCP integrations like search retrieval uh rag

02:52You also have scraping and then MCPs so that it is able to even uh introduce more plugins and tools to enhance the agent's capabilities

03:02There's human collaboration as well with human in the loop and report post editing

03:06And there's also content creation where you can create podcasts or even presentations similar to how you would with manis

03:12So there's two ways to actually install this

03:15You can either use the source code to install this by cloning the repository and then installing all the dependencies using uv npm and nvm or you can install this using docker which is way more simple

03:28You just need to have docker desktop installed

03:30You need to then change the configurations which is super simple

03:34You can use lama or you can use free models using the open router API key if you do not have the requirements on your computer to host the model

03:41This is where they showcase how you can use Olama or how you can use open router models

03:47Open router actually provides free models like deepse and you just need to simply get the API from open router and then just follow through with the installation guide on how you can set this up

03:57But due to past experiences of people commenting on my videos saying it is kind of hard for them to actually use Docker

04:03I'm going to be showcasing the actual installation method using the source code

04:07But in my opinion, I would if I have Docker running, I would follow through with the following installation guide

04:13I'll leave a link to this in the description below so that you can get started

04:16But I'm going to be now showcasing how you can easily install this using the source code

04:19So in terms of the prerequisites, make sure you have Python installed, the latest version as well as Node.js

04:25Make sure you also have UV as well as NVM and you also have, I believe, PMPM so that you can easily install the dependencies

04:34Make sure you have git installed VS Code

04:36And once you have all these prerequisites fulfilled, you can then move forward by cloning the repository

04:42It's super simple

04:42Copy this code line over here to clone the repository

04:46Open up your command prompt, whether you're on Windows or Mac OS or Linux, and then paste in this command to clone the repo

04:53Once it has finished cloning, you want to then go into the deerflow directory by typing in cd deer-flow

05:00And that is by simply typing this in within the command prompt

05:02Now, what we're going to need to do is we're going to need to configure and make sure that we add all of our API keys like Tavly, Brave Search, as well as Vulk uh engine

05:14This is where VS Code comes into play

05:17You want to make sure you click on file and open the folder that you just cloned

05:21So, the deer uh flow folder

05:21And once you have opened it up, go over to the env.example file

05:27And first things first, you want to rename this

05:29So, take out the.example

05:31And then afterwards, configure the API keys for the things that you want to use

05:36There's a couple of things that are optional

05:37So you don't want to provide it if you don't want to use it

05:40So like the TTS like uh Vulk engine, if you do not want to use TTS, you don't need to provide this API key as well as languaglow API key

05:50And just a side note, if you're using the Tavly search engine, make sure you keep that as Tavly, but if you're using duck.go, you would obviously replace this with the following other uh search engines

06:00and then provide the API key for that search engine

06:04Now, the next part is super simple

06:06Head over to the conf.ym file, and this is where you're going to be setting up your large language model

06:12So, this is where I want you to read the configuration guide

06:14Uh, this is super important because they talk about how you can switch between models and what models you can actually use

06:20You can use the ones that are OpenAI compatible models and then you can set them with the base model, the base URL, the model, and then the API key

06:28It is recommended that you use deepseek but obviously you can use O Lama models as well

06:33So these are models that you have installed locally

06:37You also have open router models that you can use

06:39Like I stated before you can use a free model and a free API key from them

06:43And it's super simple

06:45You just need to set and configure the open uh router API key and then the base model as well as the model like we did over here

06:51And then at the end provide your API key

06:53So now what I'm going to do is set this for something like deepsek

06:57And it's super simple

07:00I can just simply go back into this and then paste in my base URL, the model, which I'm going to be using V3, and then the API key that I have for this model

07:08Also, make sure you rename this and take out the example cuz that is pretty important

07:14And once that is done, click on file and then save

07:16And then you can then head back into the GitHub repository

07:20Then next, you want to install MARP ply

07:22And this is super simple

07:24If you're on uh Windows, you want to use the npm install command

07:26But if you're on uh Mac OS or Linux, you would want to use the brew command

07:32After that is installed, you can then go into the web directory

07:34If you want to use the web UI, but if you want to just use the console UI, just use the following command

07:41UV run main.py

07:41And this is where you can run it like within your command prompt

07:46But in my opinion, that does not look too good

07:48So we're going to go into the web directory

07:50And then you want to then install the dependencies for the web UI

07:53And that's by simply pasting in this command and clicking enter

07:57And once it has finished installing, go back to the repo

07:59And you want to make sure you then start up and run the backend and frontend servers with the following command

08:05And then it will open up within our local host

08:06So for Windows, I'm going to copy this command

08:10But if you're on Mac OS or Linux, copy this command

08:11And this is uh important

08:14Make sure you go back into the deerflow directory

08:15In this case, if you're on Windows, you command cd space dot dot and then use the bootstrap command

08:22Now, it looks like I'm having an issue because I cannot find UV

08:24I'm going to set the path and then I'll be back

08:28And there we go

08:28We now have our Deerflow agent running locally

08:30And you can see within our server, it is actually running

08:34And then our backend server is also running

08:36So now what we can do is simply click on get started

08:41And this will take us over to the next page where we can now send in any sort of prompts and use the deep research agent locally

08:47So now let's test this out

08:49In this case, I'm going to ask it how many times taller is the Eiffel Tower than the tallest building in the world

08:54And you can see right away, it is working on uh calculating this response

09:00It is now working on making the plan

09:02Like we had mentioned before, there's the planner agent

09:04So now we're going to ask it to start the research

09:06This is where it's going to use the Tablety API key, scour the web, and you can see it's going to then work on finding different results to get you that correct answer

09:16So, let's see what it actually ends up generating for us

09:20If you like this video and would love to support the channel, you can consider donating to my channel through the super thanks option below

09:26Or you can consider joining our private Discord where you can access multiple subscriptions to different AI tools for free on a monthly basis, plus daily AI news and exclusive content, plus a lot more

09:40And just like that, we have our deep research agent giving us an answer on the height comparison of the Eiffel Tower versus the tallest building

09:47And it did it pretty quickly

09:49It should have generated a report, but it looks like I needed to specify that within the actual prompt I had

09:56But essentially, that is your local deep research agent that replicates Manis and many of these other general AI agents we showcase on this channel

10:04Now, this is an amazing tool and there's so much more potential to this

10:08So, I'll leave all these links in the description below so that you can easily get started

10:11This is something that you can easily get locally completely for free with a free API

10:15So, I highly recommend that you take a look at this

10:19But with that thought, guys, thank you guys so much for watching

10:22Make sure you subscribe to the second channel

10:23Make sure you also join the newsletter as well as the Discord

10:27Follow me on Twitter

10:27And lastly, make sure you guys subscribe, turn on notification bell, like this video, and please take a look at our previous videos cuz there's a lot of content that you will truly benefit from

10:34But with that thought, guys, have an amazing day

10:37to spread positivity and I'll see you guys really shortly.