AMA | Behind the Launch: Sahara AI Agent Builder and AI Marketplace Launch

In this AMA, the Sahara AI team takes you behind the scenes of one of our biggest launches yet: the Sahara Agent Builder and the AI Marketplace. Joined by CEO & Co-founder Sean Ren, AI Product Lead James Costantini, and Marketing Lead Joules Barragan, we explore how these tools are lowering the barrier to AI agent creation, streamlining the path from idea to deployment, and introducing on-chain ownership for AI assets. Whether you’re a no-code creator or an experienced developer, this session breaks down the vision, the product decisions, and the future roadmap for agent monetization and marketplace growth.

Link to Full AMA: https://x.com/i/broadcasts/1PlKQMmkwZDKE

Transcript

Joules: All right, welcome, everybody. Hello, I'm Joules with the Sahara AI team. Today marks the next major unlock in AI development: The Sahara Agent Builder and the AI Marketplace are both live now in open beta. Together, they mark a turning point for anyone building or hoping to build AI agents. We're excited to dive into the details with you all today and talk about what you can expect from this launch. If you have any questions throughout this AMA, please leave them in the comments, and we'll get to as many as we can at the end. All right, let's get to it. Today, I am joined by Sahara AI CEO and co-founder, Sean Ren.

Sean: Hey everyone, I'm back.

Joules: And our AI product lead, James Costantini. 

James: Great to see you guys.

Sean: Thanks for joining.

Joules: Yeah, thank you both for joining us today. Sean, for newcomers, we just had our TGE, so we have a lot of new people interested in Sahara coming in today. Can you give us a 30-second elevator pitch? What is Sahara? How do this new agent builder and marketplace fit into our vision for building this open, equitable infrastructure for AI development?

Sean: Yeah, sure. Welcome everyone. Glad that we can get connected here again. Sahara AI is building this full-stack AI-native blockchain platform that really ensures ownership, provenance, and opportunities to monetize your knowledge through creating AI assets. We're thinking about putting your knowledge into datasets, into models, into agents, and really using this as a tool and medium to make you a content creator, to make your ideas flourish in this AI-driven future. The agent-building and registration launch that we are doing today is a critical piece of the entire bigger picture because it allows anyone—no matter if you're a techy developer or an entry-level AI creator—to use our user-friendly, relatively low-code interface to create your AI agent.

More importantly, you can register agent ownerships on-chain while collecting your ownership NFT through our registration process. So this is not just about creation but also about claiming your ownership on-chain so that it traces the provenance and your contribution if other people are going to use your model and agent down the road. To me, this opens up a whole new space about how you protect your copyrights and ownerships in this entire AI-driven future.

Joules: That's amazing. So from a product perspective, James, what exactly were we trying to unlock with this agent builder and marketplace that just hasn't been possible before? I know Sean just touched on this concept of ownership and autonomy.

James: Yeah. So Phase One of the agent builder, which launched today, makes the time-to-first-agent a one-page, four-step flow. You name it, you pick a model, you can flip on RAG or file search, and then hit deploy. The goal was to try to make it as easy as possible to go from zero to one and build agents without having to stitch together prompts, datasets, and infrastructure by hand. Our serverless breakdown makes it as easy as a few seconds to go from testing to deploying as fast as possible.

Joules: And then we have a lot of new AI-curious people in the crowd today that may be trying to develop agents for the first time. Could you briefly explain what RAG is for the audience?

James: So RAG is file search, essentially. It makes it so that any dataset within our marketplace or any files that you upload to the LLM for the agents can actually bring that data back in real-time and use it within the chat.

Joules: Awesome. Thank you. Today, building an AI agent from scratch still feels like a very gated experience. We just talked about some of the newcomers we have that might be interested in developing agents but don't have the technical know-how. You also have developers who have to go through all these fragmented tools. There's this high technical overhead. I'm curious, what are the real structural issues holding back broader participation and AI development? Sean, I'll go to you first for that one.

Sean: I think there are two important factors that are creating friction today with agent development. One is that collecting the resources to build the agents is a very fragmented experience. For example, you decide to build an agent as a personal replica of yourself that can talk to your fans on X. You need to collect a few things: your personal chat histories from different apps so your AI can mimic what you say or your style; you also need to find good models that talk like natural humans in a social setting rather than a very professional tone. You might also need to find some other good knowledge resources that represent your topics of interest, like maybe DeFi or meme coins.

All these things you need to collect from the internet, but now we provide you an AI Marketplace with all of the proprietary and open datasets listed there, where you can just browse it based on different topics and tags, and then find a dataset of your interest. You could also go through what we launched last month, this data upload and registration process, to upload your personal chat data onto your vault and then reuse that through the asset search process. So every dataset, no matter if it's on the marketplace or from your local machine, gets one simple user experience connected together so that you can use them as easily as you want. Models as well, you can find them and different proprietary and open models from the marketplace so that it's super streamlined to find and collect the resources for your agent.

The second big factor is that if you are not an advanced developer who can write a bunch of code in Python and then do all of the API calls to build an agent, you need a very nice user interface (UI) that works you through it. We make those steps very simple and natural for you to connect the dots. By accomplishing these two factors, we really lower the barrier and make it a more interesting experience for you to build an agent, like putting together a Lego, for example.

Joules: I love the Lego block example because it's literally what it is, right? You're just stacking all these things together to create something really cool. James, I'm curious, you've worked with a lot of builders. How is everything that we're building right now, everything Sean just described, different from some of the tools that are already out there?

James: Great question, Joules. Most low-code tools that are out there still expect some CLI setup, some command-level environmental variables. They need custom RAG or file search plumbing, and those hoops block non-ML or non-engineering teams. So by surfacing the model, the marketplace picks, the vault search or asset search, and the deployment all in one complete UI, we remove the problem of jumping between three different dashboards or platforms that has traditionally been killing adoption.

Joules: I love that. I'm really excited to see what kind of agents a lot of these non-devs start to build. I'm curious, what are some of the hard decisions or hidden details that make the agent builder powerful without sacrificing usability since it is this low-code, no-code experience?

James: Our single-step wizard mirrors how people think. It has the basics. You can easily choose your model, add in the data, and then you can test, all on a single screen with live token costs and cost hints in there. That makes it very easy for people to really understand how to build it. The inline asset search toggle auto-saves information from the vault, so you can easily bring that in. We've debugged everything to kind of make it so that this system is extremely fast. I think those kinds of hidden little pieces of trying to make it as easy as possible and everything can be done on a single screen is one of the main benefits of our new agent builder.

Joules: One thing that you didn't really touch on that's also on that same screen is the agent registration process. Can you talk a little bit more about that?

James: Yeah. Our agent registration process is brand new. When you go through and finally save your agent, there's a quick little screen that pops up that allows you to say, "Who owns the data that I'm doing? Who's the co-owner of this agent?" It walks you through a very simple process to take that and say, "Here's my agent. There are commercial use terms, it's free use, I want to have somebody else pay for this, or anyone can use it." It tracks all of that on-chain in an ERC-721 smart contract that is there forever. So once you make your agent and you decide you want to eventually list it on the marketplace, the registration is there so that we all know where the data came from, where the model came from, and who actually built the agent. So eventually when we have monetization, it makes it very easy to say, "This agent costs this much. Here are the different people that can get compensated for that."

Joules: Sean, a lot of this might sound really familiar because you talk about it all the time—how on-chain registration and provenance tracking are really critical to monetization. Can you go a little bit more into why that's necessary for this monetization?

Sean: I think everyone is starting to learn that agent creation or the iterations of agents is not always you building something from scratch. It's more likely that you build something on top of other people's existing assets. We're talking about if there's a great model or a great agent on the marketplace, someone else can get the license access to the agent and then contribute their good datasets into the agent's asset search functions. Or you can help modify the prompt, or you can replace that model in that agent with a better model down the road. What essentially happens is you are creating a better and newer agent based on someone else's results, and someone else's previous version of the agent should still be a contributor to your new agent. If you're the owner of the prior agent, you want your agent to be used by so many people to be iterated into a better version, so you still become the stakeholder, the co-owner, the contributor on those newer agents that are able to monetize on the marketplace. That's something we call the lineage of this agent. We want to really trace those relationships on-chain, make it transparent and auditable, and everything can eventually run autonomously on the monetization aspects. That's why the on-chain registration is super important.

Joules: Can people still use the agent builder if they don't want to register or monetize their agent?

Sean: Yes, for sure. The agent builder and the registration process are framed as two sequential steps. You first build the agents with our workflow, and then you have the option to wait until you've polished it enough before you monetize it. But the registration process is an integral part of the whole agent-building process. You create an agent, you need to claim the ownership. But this agent doesn't have to be public. You don't have to release this on the marketplace yet until you figure out how you want to do the business. So you can think about on-chain registration as a necessary step for agent creation and agent publishing on a marketplace, and monetization is the next optional step.

Joules: One question I'm seeing from the community is, when you register your agent and publish that registration on-chain, what information exactly is publicly put on the chain?

James: The owner, the datasets that are being used, the model that is being used, and if there are any co-owners, that's everything that's in. None of the prompts are actually saved on there, so people can't see publicly what the prompt is. The rest of the details of when it was created and what model is being used are all saved on-chain.

Joules: As we give more people the tools to create and deploy AI agents and eventually list them in the AI Marketplace, what responsibilities do we carry as a platform to ensure this power is used safely, ethically, and transparently?

James: We're building a lot of observability and moderation tools into our platform that is a core of what we do. We find it very important to make sure that as a platform, we're enabling the right type of development to happen. For the data marketplace, we do have a full set of moderation tools to make sure that data coming in is actually owned by who it's owned by and that it's not replicated from another dataset that other people can buy. We'll do the same for the agents as well. There will be both automated and human moderation. We'll be using LLMs to evaluate the agents that are coming in, and then our human moderation team comes from our learnings with our data services platform that we launched years ago. We have a network of people that can help grade and give quality reports to the agents and how they're being used. Our core is always to make sure that the tools are being used correctly for good uses, but also as the model developers and data providers have actually intended their technology and data to be used.

Sean: Just to add on to what James said, I think the bigger direction we are going here is for one, using some basic AI-powered tools or evaluation processes to make sure that the agents, datasets, or models published on the marketplace are meeting the basic requirements. I think the more advanced path is we still need to involve humans in some way to eyeball the models and test them. There's a very organic way that we connect our data services platform with our AI developer platform, where once a model is published on the marketplace, we will be able to automatically establish some evaluation tasks on the data services platform and incentivize people to vet the agents or models. We will then have a certificate or a verification proof on the model's utility from the reviewers. Once that whole workflow becomes almost autonomous, we will have a very scalable framework to vet these AI assets on the marketplace when they get published.

Joules: I love how interconnected our ecosystem is. It's really amazing. Speaking of the marketplace, I just want to recap some of the big things we said here. It feels like the big unlock is how end-to-end this experience really is for AI development. For builders, it's no longer having to scramble across five different tools. You can bring your own dataset or you can access open datasets in the marketplace, select a model, get compute, monitor performance, everything you need basically in one platform. Why was that level of integration such a priority, and how does it change what people can build or prototype on our platform from day one?

Sean: The biggest reason here is that monetization is always a puzzle in the existing AI development platforms out there. You have so many nice tools, low-code or no-code. You have Hugging Face, which is a great open-source community, but the biggest pain point today for developers is they don't have a good distribution channel to monetize their assets in a one-stop experience. They have to build this agent somewhere and then figure out themselves how to monetize the model, which is a very tedious and uncertain process. What we want to create is a very standard process for you to publish these agents on the marketplace, and people can explore and play with it. If they're happy about it, they can license to use it for a certain period of time. We have diverse business models to help you monetize your agents, maybe pay-as-you-go usage, maybe subscription models down the road, and so on. I think to close the whole loop of creation and then monetization is the most important design choice that we decided to go for with the whole agent builder plus AI marketplace.

Joules: Cool. And then, James, we're still in open beta. Can you talk a little bit about which features are currently available in open beta for the marketplace and the agent builders and which features are coming up?

James: For the Agent Builder, there's no marketplace to it yet, but there is still a data marketplace where you can add data in. Phase one of our agent builder is about speed and certainty—being able to get builders from idea to a fully usable agent in as short a time as possible. This includes the ability to write your system instructions and prompt. You decide if you want to turn on file search or asset search from any of our datasets. I believe there are now almost a few hundred thousand datasets in there, so there's a lot for our builders to play with. You can then register and then deploy. The deploy is another very cool feature in that most other builders out there, you can only deploy to their hardware. We have multiple different compute providers that can compete for price. So from a single place, you can go from ideation to deployment on the best or most capable servers, or the cheapest servers that fit your needs. That's the big part of this ecosystem: we're giving the choice to the builders.

Joules: I'll just add real quick that if you are looking to be one of the first people to actually monetize your assets on the AI marketplace, we do have a marketplace partner program that you can join. Be one of the first ones to get your stuff listed when that goes live, and we have a few exclusive perks as well for those partners.

Joules: Let's open up questions for the audience. What behavior or creations are you most hoping to see in the early weeks for the Agent Builder?

James: Our goal right now is to have the DeFi kind of lens as the first agent builders that we're doing. As we add MCP servers to it, we'll have direct exchange connections so that people can actually build DeFi bots, trading assistants, that type of stuff. A lot of the datasets that we have on our marketplace are around that, so we're hoping that people can use it for whatever they want to build.

Sean: Just to add on, some little teaser: we are also going to launch campaigns around agent building where we actually encourage some of the ideas that we would love to see. I can also say right now I think we're starting with social or personal replica types of agents that can really represent your own personal thoughts about things and some of the interesting people's thoughts. You can even play with the agent to change people's lives if you want, by playing some sort of live simulation experiments using the agent. More details are coming soon. Stay tuned.

Joules: Another question that came up quite a bit is, how does the agent builder and marketplace compare to other agent builders and marketplaces out there? What are the key differentiators?

James: Our main key differentiator is that we're having everything in one place. A lot of the different agent builders, even if they're no-code or low-code, require some CLI input, some programming. The benefit of ours is that it's made for both. Once you develop your agent or do rapid prototyping on our platform, you can easily take it off-platform and interact with our API. I think having everything in a single place where you can find data assets, you can find different models—we have all major models out there, all in a single place—and then the ability to quickly rapidly prototype these agents and then deploy them to multiple different compute providers is something that is unseen in the market.

Joules: Well said, James. When will your marketplace include proprietary datasets or fine-tuned models?

Sean: We are working on a whole standard operation process of introducing proprietary datasets and fine-tuned models. There will be two main channels. One: individual developers can upload datasets into their asset page and then choose to publish the datasets onto the marketplace by going through some of the vetting processes that we're going to introduce. Two: we also have what Joules just touched on, a whole partnership program that allows us to work with our partners to systematically put in a whole batch of datasets with certain business or domain focus.

Joules: As a builder, do I just create an agent and move on, or is there more I can do with it in the ecosystem?

Sean: This got touched on in the earlier discussion, but monetization is a necessary piece of the entire agent-building and monetization puzzle. As a creator with ideas, you don't have to be a hardcore coder to create agents. We're trying to make the agent builders flexible for a range of different types of developers and creators. The most important part is that we encourage you to understand the potential if you choose to publish this agent on the marketplace. People will directly use your agent, and you get revenue flowing, or people will choose to build on top of your agent, and you will also get revenue share if other people's agents get monetized down the road. So being able to educate the entire ecosystem about that new experience and going through that in a frictionless way is something we are going to focus on in the coming months.

Joules: Excellent. Here's one from a developer: "Love Sahara. No-code is cool, but one-click for devs?"

James: So we do have a working API right now. There's documentation on our website that you can use to get your API key and be able to build it out. Pretty soon in our next release, which should be sometime next month, you should be able to do rapid prototyping and then press the export code button, which brings it right into your CLI experience, and then you should be able to add to that. You can start building with direct-to-code today, but our goal with the agent builder is to help non-coders be able to build and then regular coders be able to rapidly prototype, export, and start to customize.

Joules: Awesome. I think that's the majority of the questions. Is there anything else James or Sean you guys want to shout out to the community before we head out?

James: I want to thank you guys for joining, but I can't wait to see what you guys build on our platform. Any feedback you guys have or issues you have running into it, we're there to help. I really am excited to see builders going from ideation to creation, all while capturing the basics of cost, provenance, and on-chain accountability during that process.

Sean: I'm very excited too and looking forward to all your feedback and comments on this launch. We definitely are not there yet in terms of providing you the full-fledged experience of building and monetizing agents, but we do have a very nice roadmap that we are sharing with the community. We're all here to listen to your feedback about what things you want us to prioritize to achieve something that you guys really want. I think monetization has always been the keyword there, and that's also our priority in the next launch. We're trying to give you an end-to-end monetization experience on your AI assets, starting from datasets and then to models and agents. Please stay tuned and let's build together.

Joules: Well said. We'll leave it at that. Thank you, everybody, for showing up. We'll see you in the next one. Have a good day.

Sean: Thank you, everyone.

James: Thanks, guys.