DEV Community: Mahesh

How to Create an AI Agent using Google ADK | Python SDK

Mahesh — Fri, 04 Sep 2026 07:57:58 +0000

Google ADK makes it easy for you to create an AI agent. And it supports different language SDK to make your onboarding easier. Typescript, Golang, Kotlin, java and Python. In past I covered typescript and Golang.

Today I want to explain the Python SDK based agent creation experience I had with Google ADK. I created a video for you to check out. It includes an example of Agent chat interface for you to try out.

Python makes it very easy for you to create an agent. You can also glue the agent to variety of APIs and the libraries available in the Python ecosystem.

I decided to create an agent that makes it easy for you to check the answers from the Wolfram Alpha API. And in the fallback I made it easier for the LLM to make the query and get the answer. So the entire agent keeps functioning.

You can also find the use case of your choice. Create an UI or command line utility for the agent. And that would help you build the rest of the workflow around the AI agent.

Connecting with external APIs where you want to use the accuracy, answers and the variety of the fixing of the areas that would kind of help you build agent better as well. In that context I would say Google ADK along with any agentic coding tool can build your agents lot easier.

How to Create Text to Speech with NeuTTS and Hermes Agent

Mahesh — Tue, 01 Sep 2026 15:44:43 +0000

Hermes agent pretty much rules the Text to Speech world with it's approach of supporting them out of the box. Some of them are listed in it's list of providers. NeuTTS is one of them.

NeuTTS makes it very easy for you to turn your script into the speech audio. And a lot of sample native English and other languages supported too.

I created a video that would show you how to setup the NeuTTS and convert text-to-speech in Hermes Agent.

Hermes agent has the tools commands which would show you Text to speech libraries included. And you would find the NeuTTS listed over there. You just have to choose it, there are no API keys to add.

Another alternative is to choose the NeuTTS through github URL and ask the hermes agent to use it either through python package or clone the repo and learn from the sample.

It would take some time because VENV makes it difficult for the env setup and the packages to be run that quickly. And considering how the Hermes agent runs test before it can show it is done. You have to give it some time.

You can watch my video as I explained this setup to example process. Also have some sample text file ready as that would make TTS example quicker.

That's it. You have it setup and you are free to make the TTS experiments.

How to Connect Hermes Agent to mem9 AI Memory

Mahesh — Sun, 30 Aug 2026 08:48:48 +0000

Hermes agent is pretty much open towards the memory providers and it has plenty of them listed already. I also recently covered Hermes agent and Mem0 AI Memory post.

Now I recently integrated my Hermes agent with Mem9, which is another service which supports Openclaw and Hermes agent. It's free within limits, open source and have cloud instance too.

You can check out my video for managing the Mem9 AI memory and integrating it with the Hermes Agent.

How to get started?

First go to official website of mem9. You have two options either download it as an instance or login with gmail and github and get into the dashboard.

If you choose cloud route you would have to create an API key. And that API is what you are going to share with the Hermes agent. This would be one off operation and you'd get done. Save .env file with that API key.

You can also ask the Hermes agent to pick the API key and read from mem9 docs to set the Mem9 as the memory provider.

You would require to check the command hermes tools. From there you would be able to access the memory providers. Once you set that up make sure to also command hermes agent on what to take it to the memory.

For example,

Save my frontend design choices, personal life (safe events) are stored in the memory for you to access later. Do not store it locally. Store it in attached cloud provider for AI memory.

Make sure to also get rid of memories which are not safe to share. Because at the end you don't the providers of agents and memory use that against you as a customer.

That being said, mem9 is worth exploring and saving your memories.

How to Create AI Voice Agents using ElevanLabs API

Mahesh — Fri, 28 Aug 2026 15:57:15 +0000

When it comes to voice AI agents, there are many TTS providers out there. But if you are trying to build real time speech solution and if you wish to build a service bot or AI agent with voice capabilities it's expensive.

However there are some of the providers like ElevenLabs which offers some credits to build the apps on their ecosystem.

I created a video explaining how to Create Voice AI agent using ElevenLabs API.

Here's how I start for any new voice AI agent.

Lets say you want to build voice agent for your store. First come up with data on your store. So that agent can pick the inventory, script and the data related to store.

Use langchain or Google ADK. And this framework would build a basic agent. You can use the Vercel Eve framework to build a chat agent around this product. And you can use LLM which gives decent processing for the text to voice conversion on the go.

This is the frame of your solution. You can extend this into the prompt and build the app. App is how your get agents to work. Building agent doesn't mean just writing the code on Google collab and be at that stage.

Build the implementation with an app. That's how you learn to build Voice AI agent.

How to Connect Hermes Agent to Supermemory AI Memory

Mahesh — Wed, 26 Aug 2026 06:02:42 +0000

Hermes agent is pretty decent tool it can get the a lot of user tasks done. While there are some cases where you wish to have a memory and set of instructions. So that you don't repeat your prompts. You don't repeat your everyday automations.

This is where the memory comes in. By default Hermes agent has user.md and the memory.md. These two files hold that data. But you have limited control over that. You can extend this using external memory providers like Supermemory.

I have made a process video of how to setup Supermemory with Hermes agent. Give this one a look if you quickly want to set memory with Hermes Agent.

Supermemory gives you 5$ worth of the AI memory credits. Which would be more than enough for most of the Hermes agent usage.

You can use it to store your session data, ai agent's preferences, your stack preferences, your database and other memories even including personal.

This way you can filter out memories in such way that your workflow would be easier to put up with the local memory. And you can sync your memory with the cloud as well. This is something we can work with Hermes agent pretty easily.

Hermes agent is pretty good with the memory management and using external provider gives you more freedom to keep your rules and the memories fixed and let the agent learn from it saving you time on prompts.

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

Mahesh — Sat, 22 Aug 2026 16:04:14 +0000

Google has created a framework for building an AI agent. This framework is called as a Google Agent development Kit. And it also has variety of the language SDK for the framework.

I recently covered the Typescript SDK and now I also covered the Golang SDK with the Google ADK. I made a video for you to check out the whole process.

In this particular Agent I decided to turn the framework output into the CLI agent. It made my life easier when I used this SDK and decided to create command line output to build the final agent.

In such agent development it helps if you sign up to a service where you can get the free AI inference API. Like say Kilocode or Openrouter or similar other type of the service. I also think you can use Opencode with their free models in the API.

In this tutorial I decided to use the Weather API and that API I managed to get access to certain APIs endpoint and some of the city data for the weather related output.

At the end I made this agent such that it answers most of the weather answers, and for it makes use of the LLM output where the weather API data ends. So both those options kind of contribute for the development.

I'd recommend you to check out the Google ADK and use any SDK of your choice - typescript, python and golang SDK. I feel Python and golang just are too good and quick to build the command line apps.

How To Create Text-to-Speech with Hermes Agent & Kitten TTS

Mahesh — Wed, 19 Aug 2026 16:41:03 +0000

Hermes agent is very handy when it comes to getting your everyday ai tasks done. A lot of people create Youtube videos, podcast and also variety of AI based content. In that case they are now using the AI based voice generation.

A lot of AI voice generation online is premium and in order to create the ai voice you can use the open source libraries. One such library is Kitten TTS.

Hermes agent can work with Kitten TTS for producing your text into speech. I decided to use one sample text file in the folder and then gave Hermes agent to turn that text into the AI speech.

This below video is exact process on how I used the Kitten TTS to create the sample text file into the audio.

So what was my process to do this?

Open the command line and go to Hermes tools section. And there you would find a way to select "Kitten TTS". And here you would be able to instruct the Hermes agent on usage of this model.

You can alternatively use the GitHub repo link to the Hermes agent. By giving repo you would train the Hermes agent better. You would let that agent understand how to build the flow for the building text to speech output easily.

In grand scheme of things, making any text into speech becomes easier once you have a flow with the Hermes agent. I would suggest you create a two speaker convo script with any AI model and then feed that to hermes agent in a text file. And then use Kitten TTS with native speaker voice for conversion.

If you use this flow do let me know :)

How to Create an AI Voice Agent | Fish Audio API

Mahesh — Sun, 16 Aug 2026 14:03:27 +0000

Creating voice Agent was never easier. In past a lot of provider were charging heavy. And testing the AI models with the audio was not that easy in the past. Now a days a lot of providers are offering free audio models.

Take example of Fish audio which is offering their S2 model. And you can try out the developer API for free, check out Fish audio API.

I created an Audio agent and the voice model S2 is being used in one of the agent I created. This is a shelter agent which makes it easy to use in the existing website or mobile app. You can check out the video below to see how this is being done.

How this was made?

First thing I did was created a dummy data. Which made it possible to for me to have an agent build over. For example if you connect the Vercel's Eve framework. This would require something on which the agent to be used.

Once you create dummy data and sample questions the agent would be answering. You have given a head start. And from that onwards you can ask the agent for the voice agent questions from the sample list.

Use the fish audio API and choose the female or male voice on which you want to train the agent. Once you do that you are good to go. You can even explain your process verbally and most of the AI agents would be able to build the rest.

I used Antigravity IDE and asked it to build this agent. Here I explained about dummy data. And also explained about openrouter's free models. And got Vercel's Eve framework to build the chat agent interface.

That's it. And you have the final output as explained in the video above. Make sure to experiment with your dummy data. You can even take the voice input for ai agent to listen to and then get the output in the voice in return too.

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.