DEV Community: Mahesh

How to Create an AI agent using Google ADK | Typescript SDK

Mahesh — Wed, 12 Aug 2026 14:08:45 +0000

Google has released the open source framework for creating an AI agents. This framework is called as - Agent development kit.

ADK makes it easy to create, build, deploy and debug the AI agents on the web. It is available in variety of the programming languages such as - typescript, go, python, kotlin and Java. I happen to have learned how to use the Typescript, Go and Python.

Today I want to show what I have built with the help of the Google ADK in Typescript language.

If you are interested in watching the whole journey of launching the agent with the help of Google Antigravity IDE, check the video below.

So I had this problem statement where Solar system needs a chat agent for the students. Like take questions from the students. And the agent has to fetch the content from the Wolfram Alpha. By adding LLM support on the top we can always have up to date answers.

The agent is broken down into the component. We use the Vercel's eve for the chat frontend for the agent. And then the Google ADK is connected to the wolfram alpha if the credits run out we can use the Openrouter API for the queries.

Turns out the agent worked out nicely. I can make the queries specific to the planets in the solar system. And if an off topic queries are made the agent can get us back to the Solar system topic.

I am going to try out another agent with the Python and Golang SDK's. I'd choose another different agent niche next time.

How to Connect Hermes Agent to mem0 AI Memory

Mahesh — Wed, 22 Jul 2026 15:45:32 +0000

Hermes agent has it's own memory system which is managed through 2 markdown files - user.md and memory.md.

Hermes agent also offers you an option to use other memory providers which makes it easy to store the memories locally and on Cloud platform.

Take an example of Mem0 which is one memory AI provider. This makes it easy for anyone to store memory locally or on cloud. The below video shows you how to configure mem0 for Hermes agent.

One of the interesting thing I did with this setup is I asked the Hermes agent to separate the personal data, facts, preference and the other data to filter into local and cloud.

You get generous amount of the credits that makes it easier for you to upload the data from Hermes agent to the cloud account. And you get to check the memories online in your dashboard.

For example I explained some of the preference and the personal facts and asked to put them. So the personal data remained locally and other preferences, facts etc. to the cloud dashboard.

What's so good about mem0 memory platform?

I honestly had no clue what the service was about. First time I decided to use the AI memory and happened to find the platform like this. I managed to save my data into the dashboard and get to view how hermes does the separation.

Each day I learn something new about memories in Hermes agent and I continue to document it like this through video as well as post like this.

How to Create an AI Agent using Antigravity SDK

Mahesh — Sat, 18 Jul 2026 06:56:54 +0000

Google's Antigravity is not just an IDE anymore. They have created multiple products out of it - CLI, IDE and SDK.

So now if you want to code through CLI then you download Antigravity CLI, if you want Cursor like editor then you have Antigravity 2.0 IDE and if you want to build AI agents then you use the Antigravity SDK.

SDK is accessible through Python port as of now. And you would be able to use it with the Gemini Key or through your own choice of LLM to build from this SDK.

I created a short video on how to create your own AI Agent using Antigravity SDK. It solves the basic steps to start building the AI agent.

How do you get started? It's simple.

Download the Python SDK version of the Antigravity SDK. Let your agent know what you wish to build.

For example, let's say you want step by step implementation of the AI agent building. In this you choose to build AI agent one step at a time. I created the same example in above video.

After creating step by step implementation you give a task to this new agent. And it starts working on say finding something on internet, finding data to extract, scrape some data from amazon etc.

Who should use the Antigravity SDK?

Those who build around Gemini AI and the scripts they wish to build with the Python based AI agent. This SDK makes it easy to create one or multiple agents doing your tasks in the background. For developers who are getting started on building an AI agent they should try out this SDK.

How to Create Chatbot using Gemini AI

Mahesh — Wed, 15 Jul 2026 14:34:33 +0000

Gemini AI and Antigravity IDE are turning out to be my favorite. I recently built Chatbot using the Gemini AI. And it just works awesome.

I adapted it as per the niche of the website. And put the strong walls so that users don't get to chat endlessly with it.

Instead I made sure to focus on the products, help, support and the buying process integrated into this chatbot. This has helped me stay on focus while keeping the consumers on the flow for buying product too.

I decided to create a Progress video while building this. And that is lot better to follow if you want to build something similar with it. It just works, I also shared the prompt inside the video.

Here's how I made this:

Goal of the chatbot

You need to figure out all the hooks which chatbot would require. For example you need to have the chatbot for the product purchase through bot guidance. You need it for support, buying insurance. Just think about your use case.

Limits of chatbot

Always stop your chatbot from doing the useless chat. It should focus on limited set of the options. Always build the wall first before building the garden. Give it limits on data, what it should do, what it should answer.

Triggers for Start and End

Find out what should trigger the start and what should trigger it's end. So that way the bot would be able to respond to you and your consumers. So that would also makes it easier for building your flow for how the chatbot should answer.

Based on this design your prompt. And from there Antigravity IDE along with the Gemini AI model inside would understand your use case and then it would write the code.

How to Install Hermes Agent on WSL

Mahesh — Tue, 07 Jul 2026 17:42:00 +0000

Hermes agent is a self improving AI agent. And it makes it easy for you to manage your projects, code, everyday automations, skills.

It works better under Mac and Linux. But recently they started rolling out the setup for Windows too. So both command line and desktop app are now available.

Why would you use WSL? Simple, to keep your Hermes agent experiments sandboxed into the WSL based instance of Linux. It makes breaking things, improving your workflow, testing things lot easier.

I have created a walkthrough on how to setup Hermes Agent on WSL. That too with Ubuntu Linux setup.

How to get started?

Simple. run the following query and begin the installation setup.

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

On terminal it takes a bit of time. Going through series on setup requirements. One component after another it would install the Hermes agent.

Then it would take you through the series of questions like setting up LLM provider, adding API keys, integrating the external apps and services.

At the end when all the configurations work out. You would find the Hermes Agent showing the screen something like this with installed skills, tools, plugins.

Which LLM to use? You can try out Ollama, openrouter and mostly the freeones out there. I am sure the future update would let you use any inference service with API base URL to configure.

Initially go with the setup then expand your habits on it and it would learn and adapt as per you.

How to Create an AI Agent using Vercel Eve

Mahesh — Sat, 04 Jul 2026 14:28:23 +0000

Vercel has created an AI agent development framework to support building agents. This framework works similar to Google's agent development kit that has web module.

So if you wish to create an agent that works on Vercel's stack, then Eve framework would be a good starting point. By default it has a decent UI too which makes running even simple web based agents easier.

You can also combine Google ADK with Vercel Eve and build lot more powerful and autonomous agents.

Take a look at the example below that I created using Vercel's Eve where I created AI agent with nutritional analysis from FDA.

What does this AI Agent do?

It has some pre-filled queries that answer your queries based on the respective queries.
It can make fresh queries to USDA's API and fetch the health and calories specific information.
You can also additionally use AI SDK and fetch different set of providers.

In the process I also created an ADK router that covers most of the providers from the LiteLM's list. The repository ADKrouter has the options for both Python and typescript. It's Apache licensed so feel free to use it any way you please.

Do note that there are plenty of OpenAI compatible API inference services. Take example of Openrouter which is a famous one. It can be used with the Free models from their offering to work with this agent. I did exactly that.

You can check out my video above and let me know if you find that useful.

How does the final process looks like? Agent has this sort of result I have produced.

How to Create Browser AI Agent using any LLM

Mahesh — Tue, 30 Jun 2026 14:11:48 +0000

I recently managed to get my hands on Browserbase. The platform that allows you to start a session with browser and get to extract data, manipulate browser for certain actions and build apps based on that data.

Here's what you need to build a browser AI agent.

Google Antigravity IDE
Browserbase account

You can use any agentic coding tools so it's not something you're limited to. Its just my preference. You can use other browser APIs too so don't limit yourself to a provider.

I managed to give the whole process in this video below which you can check out step by step.

First thing is make sure to sign up to the Browserbase platform. And save the API key that you can get from the dashboard.

Second thing you can do is open agentic coding tool. I am using Antigravity IDE as it is easy to use like cursor. And it has Gemini 3.5+ onwards models to work with.

Third thing is you need a prompt. You have to explain your goal to the tool.

"Create a single page app that would scrape the Google search results for certain business. And extract data and displays it in an accessible yet minimal and presentable ways.Use the browserbase API key (attach). And here are the docs to browserbase: ref1, ref2"

From here onwards Geminin would give you an implementation plan. And you also get the walkthrough once the app is generated. Check if it does what it is intended to do. Also this is just one shot implementation. So at the end you have to refine it for final touch.

Above image is the final result of my prompt. And it is pretty decent app which i restricted to a certain business type.

Like this you can create many browser agents that serve a specific purpose.

Hermes Agent Desktop App for Windows | Free Local AI

Mahesh — Fri, 05 Jun 2026 13:30:22 +0000

I started with Hermes agent for some time now. I also recently wrote about my journey with setting up Hermes agent on the Linux OS. And turns out it is going on good so far.

I learned how to set up the Hermes agent on WSL, Ubuntu Linux. So that's two different experience I have so far. I have yet to get it on Mac and do anything in automation there.

Recently I learned about Hermes agent has released the desktop. And they are competing with the Google Spark, Antigravity and Codex on desktop app front.

I covered the setup part here in the video if you are interested.

Now few days of experience so far taught me a lot about this desktop app. So I have some complaints and some good pat on the back too.

It's not native build. Considering that part they didn't even bothered about asking user consent on which directory to install. That's the sad part, it goes to C: drive and consumes space.

Default directory goes into C drive and making it fragmented and scattered. Fortunately you get to change this workspace directory.

Openrouter is pain in the app. You may come across lack of tagging for the free models. And so if you have no credits you may end up facing 404 errors on the app terminal.

Nvidia NIM Build also acts up. Like some of the models are extreme traffic on them. And so responses take forever. I don't blame the Hermes agent for this but atleast they should make it easier for us to choose our own custom APIs and we can fetch our providers there. Trying to be everything for everyone by spending resources on giving provider support often breaks the rest.

Rest you can connect to other free providers and begin the chat or the coding. I have decided not to move with it as I am kind of happy with the Cursor, Antigravity and the Devin setup so far.

If you are looking for UI version of Hermes agent then this is a good adventure to try out. But eventually coder in you would want to settle with the CLI.

Building Health Record App with Gemma 4

Mahesh — Sat, 23 May 2026 05:22:35 +0000

This is a submission for the Gemma 4 Challenge: Write About Gemma 4

When I discovered the agentic coding I ran out of ideas. Yes despite being a coder and IT experience. I just found myself frozen and can't think of solving any nearest problems at first.

That was upto the point where I saw other people building vibe coding applications one after another. And that triggered me to think of my own problems.

One of the problems I have is keeping track of health expenses and keeping track of health issues. I know there are many apps out there that lets you do that. But none of them were personal.

So I came up with a prompt.

Create an app that let's you record health issues, medicine expenses, doctor visits, health issues and symptoms, medicine consumption streak and insurance data. Use Vite framework and React, SQlite.

Sounds like very generic isn't it? No. If you check I was focused on making use of SQlite to keep my data local. And also wanted to use react to cut down the time.

I used Kilo code and hooked into the Gemma 4. Which you can check in the progress of this app through the video below for the final results.

First step was turning single prompt into an app. Second adding tests to make sure each entry into database and the retrieval for results works. Third making sure the app is usable locally.

So why not hosting on web? As this was personal project so I specifically decided to stay with the SQlite. And so keeping it offline only app. I could use watermelon or similar other database that would help it keep MVP on web.

The final result looks like this.

So where to from here? I can think of creating mobile app or maybe think of making this available through Vercel or netlify hosting. And maybe even opensource this if I could not maintain this myself for long term.

How to Setup Hermes Agent on Ubuntu Linux

Mahesh — Thu, 21 May 2026 16:41:49 +0000

This is a submission for the Hermes Agent Challenge

Hermes agent has managed to catch up attention of automation engineers and the developers in just 3 months of establishing itself. Almost everyone around me are trying to setup Hermes on their VPS or the mobile phone by setting up a Linux VM.

Earlier I thought it would require a VPS setup in order to do most of the integration like Openclaw has. Then I came across some of the YouTubers who setup Hermes agent and managed to everything including agentic chat for creating tasks and scheduled jobs to get the work done.

I couldn't just stand watching other people set things up while I just get to play it safe not getting my hands dirty. So I managed to setup the Hermes Agent on Ubuntu Linux.

Below I have documented the visual journey on How to Setup Hermes agent on Ubuntu. It was installed on Ubuntu 26.04 LTS version.

Here's what you should know.

Basically you run the following command and follow the steps. You have option to follow recommended setup or manually setup the required steps.

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

If you are new to this, best option would be go with some defaults. That would get your to the setup onboarding lot faster.

What should be the minimum hardware?

You need minimal 4GB onwards RAM on your computer. I had it on the 8GB Ubuntu VM and turns out it works smooth.

You need an API key with Openrouter or Ollama cloud. With which you can use the free models from their catalog and still get to use the Hermes agent like premium user.

Which models are good to try out?

Minimax M2.5
Kimi K2.5
Nvidia Nemotron
gpt-oss

These are often the free and low cost models you can find in most of the API services out there.

You can integrate faster if you use the web UI or the desktop shell for the Hermes agent. But that's not required you can still integrate from command line. But it becomes easier and faster if you use desktop or any Visual setup.

What you should do once you get comfortable with desktop setup of Hermes agent?

If you can afford move to VPS, some box as small as 4-6GB of RAM and 60GB SSD would be a good starting point to use the Hermes agent from the cloud. Integrate all the required tools and you would get more powerful setup to access from phone and web.

That's it. Train your Hermes agent and enjoy your Automations.