The Supabase Alternative Built Natively for AI (Powabase)

BBetter Stack
Computing/SoftwareSmall Business/StartupsInternet Technology

Transcript

00:00:00This is PowerBase. It's a Postgres database, a RAG engine, and an agentic workflow builder,
00:00:06all combined into one unified backend. It's the ultimate backend as a service for modern web apps
00:00:12that rely heavily on AI features like vector databases, RAG pipelines, or AI agentic workflows.
00:00:19So in this video, we'll take a closer look at PowerBase, see how it works, and we're going to
00:00:24test it out by building a really cool retro style AI-assisted product website. It's going to be a
00:00:29lot of fun, so let's dive into it. So PowerBase.ai, not to be confused with PowerBase.com, which is,
00:00:40I don't know what that is, is a new backend as a service platform that really is a one-stop-shop
00:00:46solution for modern AI applications. So picture this, you're building an app, but you need a database,
00:00:51maybe even a vector database, and now you decide you want to build a RAG pipeline based on your
00:00:57organization's internal documents, and now maybe you want to create a UI chatbot that can answer
00:01:02questions based on your RAG setup. You could build all of those things as standalone services,
00:01:08but then it gets messy when you have to connect them all. You need to hook your database to your RAG,
00:01:13and then build an agentic workflow for your chatbot. It all gets very complicated really fast.
00:01:19So PowerBase is meant to make your life easier by providing one unified platform for all of these
00:01:24things. Because it extends SuperBase open source foundation, it uses Postgres as the single source
00:01:30of truth for absolutely everything, and that's why you can use PG vector as your primary vector database
00:01:36as well. And because your standard relational data and your new vector embeddings live inside the exact
00:01:42same Postgres engine, they share the exact same ACID transactional safety. So if a database transaction
00:01:48rolls back, the vector update rolls back with it as well. And it also integrates an agentic workflow
00:01:54builder right into the same back end, giving you a visual node based canvas directly inside your
00:02:00dashboard. It allows you to map out deterministic guardrails and hard business rules or setting strict
00:02:06execution limits while still allowing the LLM to dynamically call tools and reason through tasks.
00:02:12So all of that sounds great, but let's test it out and see how it actually works. So once you create your
00:02:17account, you're eligible to get $20 of PowerBase credits for free once you answer a simple questionnaire.
00:02:24And once inside the dashboard, we can see that it looks very, very similar to Supabase,
00:02:29but with the additional sections dedicated to all of the AI features. So a few days ago, I found this
00:02:35super cool computer hardware product catalog from the 80s on the internet archive, which features so many
00:02:41fascinating old pieces of tech. So I decided to take this catalog and build a retro style product website
00:02:48with an AI chatbot that can give me product recommendations based on this catalog. And the
00:02:54internet archive page already has a TXT file, which includes the entire text of the PDF catalog
00:03:00scraped with OCR. So we can use this text file as the data source we will ingest in our RAG pipeline.
00:03:06So to start this project off, I just have a simple folder with a picture of an old Macintosh computer
00:03:13and a reference image of what I want the website to resemble. And then I also have the text file that
00:03:18we just downloaded. And based on PowerBase's own documentation page, they have a handy start guide
00:03:24how to use it with CloudCode. They don't have a skill yet, but if we provide the coding agent with our
00:03:29base URL and a secret key and a reference to the docs, then CloudCode can basically figure out the rest.
00:03:36They do say that they are working on a skill for this project, but as of this recording, the skill is not yet
00:03:41available. So we'll just stick to their manual instructions. And for the prompt, I'm basically telling
00:03:47CloudCode that I want to build a retro product shop with an AI chatbot, and I'm asking it to use the
00:03:53reference images for design. And I have also provided it with a base URL and the secret key in a separate
00:04:00environment file. And lastly, I provide it with a URL to PowerBase's docs, just in case,
00:04:05and let CloudCode do its thing. And a few minutes later, we see that CloudCode has successfully
00:04:11extracted the catalog data from our source. And it has also created a knowledge base of our catalog.
00:04:18And a few moments later, we see that CloudCode has now successfully completed the task.
00:04:23And not only that, but it also did a test run with our AI chatbot to see if everything is working
00:04:28properly. And we can also see that updated here on the PowerBase dashboard. So everything is looking good.
00:04:35So now let's see how the website actually looks. Oh my God, this looks so good.
00:04:42Wow. It's honestly a lot better than I expected. Look at that retro theme. CloudCode has done such
00:04:48a great job with the design, I have to say. But the most important thing is the chatbot
00:04:53or the clerk, as it is called here. And as we can see here, it gives us some sample prompts we can ask.
00:05:00So let's try to see what it recommends for storing 300 floppy disks. And would you look at that? It gives
00:05:06us a very nice detailed answer. And as we can see here, it recommends either a storage binder or a plastic
00:05:13disk box. And it even lists the prices and everything. So that is pretty cool. By the way,
00:05:19I noticed 30% of our audience is Gen Z. So let me know in the comments if you know what a floppy disk is.
00:05:26All right. Let's try a custom query now. I want a strong enough computer at home that can play
00:05:33Pac-Man. What can you recommend and what are the prices? And look at that. This is interesting. It says
00:05:39it cannot find any reference to Pac-Man in their catalog. So this is a common characteristic of RAG
00:05:45engines where it tries to stay strictly within the bounds of the context of the data given,
00:05:50which is a very good thing. But it does mention a similar game that it did find in the catalog,
00:05:55which is similar to Pac-Man. And then it finally gives us these nice recommendations for gaming
00:06:01computers. And this is so funny. If gaming is your main goal, but budget matters, then we can choose
00:06:08the Interact family computer for 500 bucks. But if you want the best graphics and sound,
00:06:14then you got to go with Texas Instruments TI-99. It's a showstopper and even comes with its own 13-inch
00:06:22color monitor. Color monitor, yo. Let's go. Shut up and take my money. So the website looks absolutely
00:06:30stunning and the chatbot is working as expected. So I'm pretty satisfied with this result. One last thing
00:06:35I want to check is the power base dashboard. And here in the run section, we can see all of the sessions
00:06:42our chatbot has conducted with all the detailed responses. So this is a good way to keep track of
00:06:48how your users are actually using the chatbot. So there you have it, folks. That is power base in a
00:06:53nutshell. Honestly, I'm super impressed with this platform. It has so many cool AI features and the
00:06:59setup was so easy and the rag pipeline worked seamlessly. So I would say this is a perfect backend as a service.
00:07:07If you need to get an MVP version of your AI app out there as quickly as possible. Power base basically
00:07:13provides you with all the necessary tools so you don't have to worry about the plumbing side of hooking
00:07:18everything together. So well done, power base. But what do you folks think about power base? Have you tried it?
00:07:24Will you use it? Let us know in the comment section down below. And folks, if you like these types of
00:07:29technical breakdowns, let me know by smashing that like button underneath the video. And also don't
00:07:34forget to subscribe to our channel. This has been Andres from Betterstack and I will see you in the next videos.

Key Takeaway

PowerBase unifies Postgres-based relational storage, vector databases for RAG pipelines, and agentic workflow builders into one platform to eliminate the complexity of connecting standalone AI services.

Highlights

  • PowerBase combines a Postgres database, a RAG engine, and an agentic workflow builder into a single backend service.

  • Data stored in Postgres and vector embeddings share identical ACID transactional safety, ensuring both rollback if a database transaction fails.

  • A visual node-based canvas within the dashboard allows developers to set deterministic guardrails and execution limits for AI agents.

  • New users receive $20 in platform credits upon completing a simple questionnaire.

  • The platform enables a RAG engine to stay strictly within the bounds of provided context, defaulting to similar alternatives when specific information is missing from the database.

Timeline

Unified AI Backend Architecture

  • PowerBase extends the open-source foundation of Supabase to provide a single platform for AI applications.
  • Postgres acts as the single source of truth for both standard relational data and vector embeddings.
  • A node-based visual canvas facilitates the creation of agentic workflows with specific business rules.

The platform addresses the difficulty of building modern AI applications that require separate databases, vector stores, and RAG pipelines. By integrating these components into one backend, developers avoid the complexities of managing disconnected services. ACID transactional safety ensures data integrity, as updates to vector embeddings and relational data occur within the same database engine.

Building a Retro-Themed AI Product Catalog

  • A 1980s computer hardware catalog text file serves as the ingestion source for the RAG pipeline.
  • Coding agents can integrate with PowerBase by utilizing a base URL, secret key, and documentation references.
  • The dashboard displays session data for all chatbot interactions to track user engagement.

To test the platform, an AI-assisted website was constructed using a historic catalog scraped via OCR. A coding assistant managed the integration by leveraging the provided environment variables and API keys to connect to the PowerBase infrastructure. The system automatically ingested the data into a knowledge base and configured the AI chatbot.

Chatbot Performance and Results

  • The RAG engine successfully provided product recommendations and pricing details based on the catalog.
  • The chatbot maintained strict context adherence, informing the user when a specific term like 'Pac-Man' was absent from the dataset.
  • The platform offers a detailed 'run' section to audit individual chatbot sessions and responses.

The resulting website featured a functional chatbot that provided specific, pricing-backed recommendations for hardware needs. When queried for items not in the database, the model identified related alternatives rather than hallucinating external information. The dashboard provided visibility into these interactions, confirming the viability of the service for rapid deployment of AI-based applications.

Community Posts

No posts yet. Be the first to write about this video!

Write about this video