Supabase Vector Reviews

I use Supabase Vector to build semantic search functionality for all the client websites that we handle, and it allows us to search by meaning rather than the exact keyword. As a team, we mainly use it for semantic search functionality, and our clients find it helpful too.

One of my clients had an e-commerce website for a fashion store, and customers were always frustrated because the traditional keyword search did not work really well. For example, if a user searched for summer clothes for the beach, they only got products that had exact keywords, but searching for light clothing for vacation would show completely different results. Many customers were frustrated about this problem and handed it over to us. We built a semantic search feature using Supabase Vector. I stored embeddings by creating vector embeddings for all 500 products using OpenAI's embedding API. Each product had the title, description, category, and tags, which we converted to vector format and stored in Supabase Vector. When a user types a query, the query is converted into a vector, and Supabase Vector finds the most similar products using similarity search and returns the results ranked by relevance. Previously, 50% of searches returned no results or irrelevant products, but we changed that to 90% of searches returning relevant results. This was a very significant improvement that our client did not expect from us, and it was only possible because we implemented that semantic search functionality on the website. Supabase Vector understood that walking shoes and hiking boots are related even though they are different words, and customers could now find whatever they needed in a more natural way, searching with natural language and getting the things they need. It was a great leap for that website, and it worked really well.

I have another project that I worked on where I worked with a client who had a large content management system with over 10,000 blog posts and articles. They wanted me to build an AI-powered chatbot that could answer user questions based on their content. I built a RAG and created vector embeddings for all 10,000 articles and stored them in Supabase Vector. When the user asks a question such as how to optimize a website for SEO, the query is converted to a vector and Supabase Vector finds the most relevant articles, mostly the top five. An AI model like an LLM reads those articles and generates a helpful answer that is sent back to the user. The answer accuracy was about 85%, and the users were satisfied. It saved a lot of the client's time, and the client specifically mentioned to me that they saved about $2,000 per month on support staff hours. This was a significant win for our agency, and all we did was store all the vector embeddings efficiently, and the similarity search was faster, scaling really well. The integration with other Supabase Vector features like authentication and a database made the whole system easier to build. Both projects were great successes, and Supabase Vector has been essentially the root cause for these successes.

We supported a small tech company with a knowledge base of over 5,000 support articles where users needed to search for help articles relevant to their queries, filtering for articles from the last six months and from specific product categories such as mobile app or desktop app. This posed a major challenge for us. Without hybrid search, if a user searched for help on resetting a password in the mobile app, they would see all relevant articles across the entire database, potentially including outdated articles. We implemented a hybrid approach using SQL and semantic search. We wrote a query for the semantic search to find similar articles for resetting passwords in the mobile app while the SQL filters ensured that we only returned articles from the mobile app category created in the last six months. This approach successfully retrieved five relevant articles with an accuracy of almost 95%. Users found it very useful and received the right answers. Narrowing down specific details was only possible with this hybrid search.

Based on my experience with Supabase Vector, I personally feel that Unified Data Storage, where we store vector embeddings right next to our relational data in PostgreSQL, is one of the top features I appreciate. Additionally, there is another feature in Supabase Vector, which is SQL and Vector Hybrid Search, allowing us to combine traditional SQL queries with semantic search effortlessly. This is absolutely amazing because it lets us filter by metadata while doing similarity search. Another feature that stands out is the fast similarity search with multiple index types. Supabase Vector supports IVFFlat and HNSW indexes for fast approximate nearby neighbor search, giving us, as an agency, the ability to reduce response time to under 100 milliseconds, even with over 25,000 articles, which is a killer feature. The built-in authentication and security, with all the RLS policies applied automatically to vector queries, gives us trust with all the clients we work with. Supabase Vector's seamless integration with all AI tools, such as OpenAI, Hugging Face, LangChain, and Amazon SageMaker, is a significant edge for us. The free tier of 500 MB database with vector support included is a real help that Supabase Vector provides us. Additionally, we appreciate the auto-generated REST API, as we do not need to build any custom API for vector queries, making our development much faster. These are the features that are currently on my mind.

Supabase Vector has had a significant positive impact on the agency I work for by increasing client delivery. We built semantic search and AI chatbot features for over eight client projects in the past year, and these features are now standard offerings I provide to new clients as well. Revenue has increased by about 20% to 30% in value, and we reduced development time for semantic search projects from seven to ten days down to just two to three days because the auto-generated API and unified storage save us a lot of hours on back-end work. I saved over 20 hours of development time for one client project using Supabase Vector, which is a significant gain. Additionally, we have greatly improved client results, such as a 30% increase in search-to-purchase conversions for the e-commerce client I mentioned, which was a big win. There was also a 40% reduction in customer support questions after building the AI chatbot for the content management client I discussed, allowing the client to save about $1,500 to $2,000 per month on support staff. We have seen substantial improvements in search accuracy and scaled from handling over 10,000 articles to more than 25,000 articles, making scalability another significant win. These are all the impacts that we have experienced in the organization, thanks to Supabase Vector.

Adapting to Supabase Vector was relatively smooth, but there was definitely a moderate learning curve at the start. The SQL foundation, REST API, documentation, and integration with all the AI tools made it easier. However, understanding embeddings, index types, similarity metrics, and how SQL and vector hybrid queries work, as well as the RLS policies for vectors, required some time to learn.

I do not have many things to point out, but a couple of areas for improvement come to mind. For index optimization guidance, clearer instructions on when to use IVFFlat versus HNSW indexes would be helpful. Additionally, having a built-in embedding generation capability would simplify the workflow, as currently, I use external services such as OpenAI or Hugging Face for that purpose.

I have been using Supabase Vector for about one and a half years now.

Supabase Vector is absolutely stable for all the use cases I have delivered thus far.

Supabase Vector is highly scalable for small to medium to large scale applications. The free tier itself supports about 1 to 2 million vectors, and I am storing over 1.6 million embeddings across eight client projects with varying performance levels in the free tier. I believe if I move to the Pro plan, it will offer even higher storage and compute capability. I find Supabase Vector's scalability absolutely amazing and it scales exceptionally well for various application sizes.

The customer support has been excellent for my use as a new agency owner. Community support from helpful developers and engineers provides fast responses on GitHub issues and community forums. I have not encountered significant issues with customer support. I would rate the customer support a 9 out of 10.

I did use a couple of other solutions before Supabase Vector, including manual data extraction with spreadsheets and separate data pulls from various APIs. I used Pinecone for one project and MongoDB with Atlas search for another. I switched to Supabase Vector because the manual processes were time-consuming, lacked real vector search capabilities, and the accuracy of results was poor. Transitioning to Supabase Vector enhanced search accuracy while providing unified storage without sync issues, proving to be more cost-efficient.

I have definitely seen a significant return on investment from Supabase Vector, saving about $1,000 to $1,500 per client project. Across eight client projects, I saved approximately $4,000 to $8,000 a month, and for the AI chatbot, I saved $2,000 by eliminating support staff hours. Development time has drastically reduced, as the automatically generated REST APIs minimize the major development work. Instead of handling projects from seven to ten days, we now complete them in just two to three days. For the eight projects I have managed, we save about 160 to 240 total hours. There has also been a 30% increase in search-to-purchase conversions for our clients, along with 90% to 95% accuracy with the similarity search. Overall, development has accelerated, and the client value we deliver improves, leading to increased profits for our organization.

My experience with pricing, setup costs, and licensing has been very positive and cost-effective, as I utilize the free tier, which includes a 500 MB database with vector support at no cost, allowing support for millions of embeddings. This greatly benefits agencies like us, as it is free forever and does not require payment for a separate vector database service, covering most of my client projects without incurring costs. We have recently switched to a paid plan, the Pro plan for $25 a month that provides more storage, compute, and features. The zero setup cost and lack of upfront infrastructure costs, technical setup, and database configuration are important features that help us scale our services at a lower cost, making my experience with pricing and setup costs extremely positive.

I did evaluate other options such as Pinecone, Weaviate, and Qdrant before deciding on Supabase Vector. I ultimately switched because of the comprehensive features it offers.

I would rate Supabase Vector an 8.5 overall. I would round that to a 9 because all the strongest features such as unified storage, hybrid search, and fast similarity search that we usually use are killer features. Supabase Vector saves a lot of development time for our agency and is essential in terms of semantic search and RAG systems, which gives me about six points. The remaining three points come from the free tier supporting millions of embeddings at no cost, which is a significant benefit for all agencies. I decrease one point due to minor documentation gaps for advanced features and the absence of built-in embedding generation, leading me to choose a score of nine. My overall review rating for Supabase Vector is 9 out of 10.

My advice to others looking to use Supabase Vector is to start with a project, build things, and get to know the features within Supabase Vector, especially exploring semantic search. This approach provides ample opportunity to understand how Supabase Vector works while working with substantial amounts of data. Therefore, I encourage everyone to begin with a small project, manage a use case, and once it delivers returns, consider transitioning to the Pro plan if necessary. It is advisable to switch from Pinecone or MongoDB Atlas search to Supabase Vector for an extensive hands-on experience that significantly reduces development time.

I use Supabase Vector's Postgres for my projects, specifically vector-based projects for database usage. I also use it for my personal use on my personal apps that I build on the side. I work on an AI-based financial app, so I use Supabase Vector for that. The vector builds right into the top of that with pg_vector. The main use case that I use is building AI features, specifically the RAG pipelines, which are really augmented generation pipelines. I store text chunks and their embeddings for documentation and a knowledge base directly into the Postgres database. It lets me do search algorithms and regular relational queries, which is super convenient.

In my most recent project, I've been using Supabase Vector for RAG-based pipeline generation. I store all my text chunks and their embeddings for documentation and knowledge as a knowledge base. I created a knowledge base in the Postgres database and use this RAG pipeline to run a search algorithm on top of the text-based database that I have created. So if a user gives a prompt, it will be easier for it to associate with the pre-existing text and figure out the best results for the user's search.

As of now, that's the biggest thing I've been using Supabase Vector for lately.

The pros of using the real-time features and authentication in Supabase Vector have been evident. It's been a quick setup compared to other databases such as Pinecone, which can be more cumbersome. With Pinecone, I have to manage a completely separate service, sync data in, and deal with extra costs and latencies. Supabase Vector keeps everything in one Postgres database, so I can join vector search results with my regular relational data table easily. It's a huge advantage for most apps.

One thing I really appreciate about Supabase Vector is the backend capabilities. The Postgres changes over WebSockets allow you to subscribe to insert, update, and delete events on your table, including vector embeddings. For example, if a new document gets added to the embeddings generated, I can listen for those changes in real-time on the front end. The UI can instantly show updated search results or notify users when fresh content becomes available for semantic search. It works really well for collaborative apps such as dashboards or AI features where data is constantly updated.

The best feature of Supabase Vector is definitely the Postgres. I already know SQL, so querying vectors with cosine simplicity and Euclidean distance feels natural. No new query language or learning is required. Even the setup is really easy; you just go to the dashboard, add a vector column there, and you're good to go. That's the best feature I've felt so far for Supabase Vector. The cost-efficiency, because everything lives in one managed database, helps with the cost as well. The performance has been solid so far, and the Supabase Vector client library makes installing, embedding, and searching really easy. I also appreciate the real-time features and the authentication specificity of Supabase Vector, which I feel is solid and secure.

I initially thought it might take at least two to three months to set up a pipeline and start working. However, using Supabase Vector, I was able to set it up in a couple of days, got it validated, and started working on building up the algorithm on top of that. I am building a Python API suite on top of Postgres, so I am already refining my algorithms and searches on my databases to provide better results. I have become much more efficient than I initially estimated with Supabase Vector, and it is helping immensely.

A vector database might edge out on raw performance, but for 90% of the projects, Supabase Vector is more than enough and much less hassle than any other vector database. The time saved is literally close to 75%, which is a significant margin of time I've saved using this. The performance is pretty solid. I don't have to deal with the extra costs associated with setting up an additional environment such as Pinecone; everything is in one place, allowing me to manage costs effectively.

For advanced production use with millions of embeddings, it's necessary to be thoughtful about indexing such as HNSW or IVFFlat and hardware sizing to keep query times fast. It's not as turnkey as some other vector databases for massive scale out of the box, but the Supabase Vector team is improving documentation and performance tuning guidance. It would also be nice to have some more built-in embedding generation helpers or Titan integration with popular embedding models.

That's the one con I've noticed. I understand it's hard to integrate new toolkits with existing ones quickly after launch, but I think the Supabase Vector team is doing a commendable job. However, I sometimes worry about hardware sizing when scaling a lot of data to keep query times fast.

I've been using Supabase Vector for over one year.

Supabase Vector seems to be quite stable so far. In all my testing, it has never crashed, nor have I encountered major issues. Any initial issues I had were due to my code and internal bugs, but nothing has stemmed from the Supabase Vector side.

Supabase Vector appears to be scaling well. I began with less data and have increased it to around 10 GB now, mostly RAG data. I have not faced any issues up to this point regarding scaling but am yet to expand the data load to continue testing.

The customer support for Supabase Vector has been good. I find their documentation very helpful, and I have not had to contact customer support directly since most information is available there. Even Stack Overflow has provided a wealth of data for addressing bugs and challenges I faced during development.

Previously, I ran Postgres on top of AWS RDS, but that setup became expensive. Establishing my RAG-based vector system on AWS proved quite challenging and costly, particularly trying to connect Bedrock to Postgres within AWS. This led me to switch to Supabase Vector, which was much cheaper in that context.

I initially thought it might take at least two to three months to set up a pipeline and start working. However, using Supabase Vector, I was able to set it up in a couple of days, got it validated, and started working on building up the algorithm on top of that.

Currently, I am managing with fewer data engineers by using Supabase Vector. The time savings are notable; I expected to take two months to set up, but I accomplished it in just two days. That's over 50 days saved. Although I haven't seen major returns on investment yet, I look forward to realizing more profits down the line.

I am currently on a pay-as-you-go plan and am even considering upgrading to the pro version.

It's generally cheaper than integrated vector databases. I pay for the underlying Postgres component, storage, plus any row read or write for the vector, with no surprise charges for API calls in many cases. It scales well without inflating costs for simpler and medium-sized projects.

I evaluated other options, primarily Pinecone. The setup felt very challenging. I also considered Weaviate, which also had a heavy setup. By the time I was going through the setup information for Pinecone or Weaviate, I managed to set up Supabase Vector in a much faster timeframe and start running my projects.

For moderate usage, Supabase Vector is quite reasonable. I would suggest that others consider using Supabase Vector, as I believe it offers significant benefits. I would rate this product a 9 out of 10.

Supabase Vector serves as the retrieval layer in my RAG application, where we store embeddings of our documentation. When a user asks, 'How do I reset my password?' we generate an embedding for the query and perform a similarity search in Supabase Vector to retrieve the most relevant documentation, passing that context to the LLM for an accurate response. This is primarily for an AI knowledge base, and while there are different use cases, this is the most common one.

In my workflow, Supabase Vector acts as the semantic retrieval layer. When new documents or product descriptions are added, I generate embeddings using an embedding model and store them in Supabase Vector along with the original data. When a user submits a natural language query, I generate an embedding for that query, perform a similarity search in Supabase Vector to retrieve the most relevant records, and then pass those results to the particular LLM, ensuring the model answers based on our own data rather than relying solely on its pre-trained knowledge. This workflow is primarily related to the RAG application.

Supabase Vector offers many good features, beginning with semantic search. Another excellent feature is the integration of SQL with vector search, making it great for RAG, scalable indexing, and a simpler architecture. These are the main best features I find in Supabase Vector.

The SQL plus vector search combination has helped me significantly because it is easy to set up with PostgreSQL as it removes the need to choose between structured filtering and semantic search. By using both together in one query pipeline, vector search is smart, but not precise enough alone. Therefore, combining SQL for filtering first, and then applying vector search on the filtered set enhances the relevance, eliminating AI hallucinations and improving user intent handling.

Additionally, a key feature worth mentioning is recommendation systems. For instance, if a user says, 'Sweet vanilla perfume for winter,' SQL filters that database for winter and unisex requests, while the vector search finds vanilla-based scents with warm, sweet profiles, combining both signals effectively.

Supabase Vector positively impacts my organization through major improvements in search relevance, faster development speed, a simpler architecture, and reduced engineering overhead, with better performance in real queries due to the hybrid SQL plus vector search. This results in faster response times, lower database load, and more predictable queries, which together create practical metric improvements.

There are a couple of minor improvements I suggest regarding Supabase Vector, such as enhancing built-in hybrid search support, which currently requires manual query writing and ranking logic management. A natively integrated hybrid ranking system would greatly improve functionality, alongside better large-scale performance tuning, automated indexing, and improved observability tools.

Regarding documentation improvement, the current Supabase Vector documentation serves basic setup and simple examples well, but it could be enhanced with more end-to-end production architectures, real-world patterns, and better examples of large-scale applications such as hybrid search re-ranking pipelines.

As for additional improvements, better query debugging tools and built-in evaluation toolkits for vector search would be incredibly helpful for developers, as would an automatic embedding refresh detection feature.

I have been using Supabase Vector for the past year, and I have used it for most of my work which aligns with the Vector DB, particularly when I want to handle anything related to RAG.

Overall, Supabase Vector exhibits high stability despite occasional performance variability under load, and while it scales well for small to medium workloads, achieving the best performance at higher scales depends largely on optimization of queries and indexes.

My experience with Supabase Vector's performance at scale indicates that it operates excellently at small scales, but as we move toward larger scales, such as multi-million vectors, it requires careful engineering to maintain predictable performance, with challenges such as query plan complexity and latency variance.

We have not had any interaction with Supabase support, as we have not faced any issues that required us to raise a ticket.

Before adopting Supabase Vector, we used a separate vector database solution, such as Pinecone, which introduced complications with data synchronization and increased latency due to the need for maintenance of sync jobs and additional compute resources, driving our decision to switch.

Integrating Supabase Vector into our existing systems is not complex; it was relatively straightforward due to its direct build on PostgreSQL, fitting naturally into our workflows without needing a separate vector service setup.

The most challenging part of implementing Supabase Vector was designing an effective embedding strategy, as that presents difficulties common to all vector databases. Additionally, chunking and structuring data for heavy use cases posed challenges, particularly around embedding consistency during product data updates.

I am not certain how Supabase Vector resources got allocated since we have a dedicated team handling purchases, and they may have acquired them from the AWS Marketplace or another available marketplace.

I have seen a return on investment from using Supabase Vector, particularly through reduced development efforts, improved search efficiency, and better user engagement outcomes. The impact includes faster feature delivery and lower infrastructure costs.

We notice significant improvements when measuring metrics such as average response times, which have shifted from 800 milliseconds to 2.5 seconds down to around 200 milliseconds to 800 milliseconds, with click-through rates for recommendations improving by 45 to 70%. Additionally, search refinement has reduced redundant searches by 20 to 35%, and recommendation accuracy has improved typically by 25 to 60% for top results.

I do not have all the specific data on pricing and licensing since we are not responsible for those decisions, but I know that Supabase Vector does not have a separate licensing model, and the pricing is primarily based on the overarching Supabase database tiers.

We evaluated options such as Pinecone, which is a leading managed vector database recognized for strong performance and production-readiness. However, we did not choose it because the separate system management introduced data synchronization complexities and additional costs.

When selecting Supabase Vector over other solutions, the main criteria were simplicity in architecture, tight integration with relational data through SQL plus vector capabilities, strong security and governance features, and faster development speed.

My advice for others considering Supabase Vector is to start simple and avoid over-architecting early on. It is crucial to design your data model carefully and always use SQL filtering first before vector search to enhance results.

Overall, Supabase Vector is a good product that needs some improvements, but it has many redeeming qualities. It is easy to integrate, especially if you are already using Supabase for your database schema. I encourage others to try Supabase Vector as it proves to be a reliable tool. I would rate this product a 9 out of 10.

At the initial stage, I used Supabase Vector in building AI chatbots and RAG agents. I used it as the vector store where I input knowledge bases and attach it to the RAG agent that acts as customer support for businesses. At present, I have started using it for web-coded apps. I basically use it as a back-end service where I can have a vector store, a back-end that is attached to the front end of the app and then I have something that serves as an authentication space for the app that provides authentication and authorization for the app users.

Supabase Vector works fine because for one of the apps I have used it for, it has had over 200 users and it actually worked fine. I have not had any issue to transfer to any other service at the moment. I have never thought of scaling or transferring to something else for a back-end because it serves the purpose of what I have used it for. I have basically used it for small teams, not large teams that need to cover thousands of users. Since it serves the purpose, I have never thought of scaling to something much more that will cover a wider range.

I appreciate Supabase Vector because it has RLS that allows its users to have complete access to just their data. With Supabase Vector, you have access to your data because of the RLS feature. You do not get to have access to other features and the database has structured tables, structured rows and columns. If I were building an app from scratch, I would have to build this back-end from scratch. However, with Supabase Vector, I do not have to do that. It serves that purpose. All I need to do is attach it with the front end of my website.

When I use it for RAG agents, instead of searching for a particular word for my RAG agent, it searches for related meanings. If I ask how to optimize customer support or how to make customer support faster, Supabase Vector will search for related meanings concerning these questions. It will give me answers relating to search bots because that would be related to customer support and helping to optimize the customer support experience. It will also give me related meanings that have to do with customer support. It is not searching for that particular keyword, but rather searching for all the related meanings. Because of this experience, it is much easier to use Supabase Vector for my work.

Supabase Vector serves the functions for me. However, they could work on debugging the RLS, which is the row level security. It is actually great, but when a policy blocks access, it can take some time to figure out what exactly is happening and why. If they could make the debugging process clearer to prevent the error messages, that will make development faster for web-coded apps.

I think the lagging part also needs to be addressed. When the website goes down, the lagging part needs to be resolved. When you have attached a front-end app and your users are using it and then the back-end is not in sync or lagging at the moment, it usually affects the front end of the app because the app will not be able to function to its maximum expectation. If that could also be sorted, then everything will be in sync.

I have been using Supabase Vector for three years now.

I have not really done much of the maintenance tasks on Supabase Vector. At the moment, I have not spotted any downsides because Supabase Vector actually serves what I used it for. I basically use it for my web-coded apps and for the RAG agent and it does all of the needs that I want it to do for my project and for my client's project. At this moment, I have not seen a downside for Supabase Vector.

Supabase Vector works fine because for one of the apps I have used it for, it has had over 200 users and it actually worked fine. I have not had any issue to transfer to any other service at the moment. I have never thought of scaling or transferring to something else for a back-end because it serves the purpose of what I have used it for. I have basically used it for small teams, not large teams that need to cover thousands of users. Since it serves the purpose, I have never thought of scaling to something much more that will cover a wider range.

I have been able to figure out my issues on my own. When I have an issue, I tend to reach out to either a community of people where we do automations. If I have an issue, I could drop it in the community and somebody is always there to guide me. Somebody who may have gone through that challenge before is always there to help. I have always been able to solve it out with the help of my community or sometimes YouTube, but I have never had to contact support. Over time, I have gone through that stage where I even have to contact anybody because I am already used to Supabase Vector because I have used it for a long time.

I use Pinecone for my RAG agent. I have used Pinecone before. I have used Firebase, but I did not use it as much. I did not use Firebase as much because we were comparing apps. When I got into automation, we were comparing what apps would serve better for us. I used Firebase, but I did not go into it as much. When I got to know about Supabase Vector, I fell in love with it since it worked for what I was using it for, and I just stuck to it.

Supabase Vector is very easy to deploy. You do not have to go through a lot of stress to deploy it and you do not even need to have technical knowledge to deploy Supabase Vector. If you go on YouTube, and you are a newbie trying to deploy Supabase Vector, you can get on YouTube and since it is open source, it makes it even easier to deploy. You can go on YouTube, check out the videos or a tutor can guide you on how to deploy. It is easier. You do not need so much technical knowledge for you to actually deploy Supabase Vector.

I think the pricing for Supabase Vector is perfect because if you need it for your business, you should be able to pay for it. It does not have to come free because you are getting value from using it. Whatever gives you value, you have to spend on it. For now, I think the pricing is perfect because every business person can afford it and a developer can afford that price. It is at a good range at the moment.

I use Supabase Vector for semantic duplicate detection. The app allows citizens to submit and vote on political proposals, and we do not want twenty nearly identical versions of the same idea. When a user submits a proposal, we generate an embedding of the title and description, run a vector similarity search against existing proposals, and if anything scores above a set threshold, we show those matches to the user so they can vote on an existing item or an existing element instead of creating a duplicate.

The approach using Supabase Vector works really well for me. The app does not have a huge amount of users, but some of them have used that feature and avoided duplication. I'm really happy with that. Although it was a proof of concept at the beginning, when I launched that application in real life and saw users use that feature the way I was expecting, it was really encouraging.

The feature regarding duplicate detection is basic and simple, and I was expecting the user to avoid duplication in the database. It helps to reduce the number of elements that create noise and bad use cases or bad elements. It was really nice to use that implementation and that technology in my app.

I think the best feature Supabase Vector offers, or the most valuable thing I've noticed, is that pg_vector comes built in straight out of the box. I did not have to stand up or maintain a separate vector database. Getting one source of truth or one point that I can keep everything together is the main thing. The embeddings live in the same PostgreSQL as my relational data, so I can join them with metadata, apply row-level security, and keep one source of truth. I think that is the key part and most valuable for me in that regard.

Having everything in one place, embeddings and relational data together, gives me full control through SQL. I can write my own RPC functions from the similarity search using the cosine distance operator, all inside the same database. That gives me control over the threshold, the columns, the return data, and the queries I'm using. Getting all these things together in the same database is a huge advance and it was a very neat implementation of the feature I wanted to build.

I was using Supabase, so not only getting the database but also the authentication, the vectors, and the connections with webhooks to connect everything together, because the app natively used webhooks to connect with n8n. That flow was in charge of connecting with the database on Supabase. Being able to create that flow was amazing to get all that information running from one part to another and create that flow that connects everything. I was also able to launch Firebase notifications straight away to all users using n8n, so it was amazing to manage these complex features with those technologies.

I had problems integrating the vector directly into Supabase, so I had to use Google Vertex to generate the embeddings and the information I needed in the database. It would be nice if all of this could be integrated all in one place with Supabase. I would prefer not to have to use a different or external tool to create these embeddings. It would be nice to have everything integrated in the same way. Apart from that, I think that is one of the cons I found, but it is basic.

That was my main concern in that regard. My advice for others looking into using Supabase Vector is to take into account how you are going to create the embeddings, as it is not an option implemented straight away in Supabase. You will need to handle the embeddings with an external tool, so make sure to consider how these integrations are going to be made.

I have used Supabase Vector for approximately six months.

From my experience, Supabase Vector is stable. There have been no failures, and the app is up and running, so I have no concerns in that regard.

Regarding Supabase Vector's scalability, the app is still in a very basic stage, so scalability has not been an issue at the moment. The app is getting users, but it is not having thousands of users creating problems with scalability. Therefore, I do not have enough information to answer this question properly.

Customer support is handled using emails at the moment, and since I am not using customer support from Supabase, I suppose it does not have any problem or concern in that regard.

At the beginning, I was trying to build that part on my own in my VPS, trying to build the database and everything together. When I realized that this implementation was much more complex than I thought, that is when I decided to move to Supabase. The integration problem was how to connect this complex flow to detect duplicities between the app and Supabase. I decided to integrate n8n to control that part of the flow and determine when a user was creating a duplication and needed to return to the app, or if the proposal was new, go straight away to the database to create embeddings and store everything.

I have always worked on Firebase, but I was tired of using the Firestore database, which did not cover my needs for this project and the features I wanted to implement. I realized that due to the lack of implementation or technology in Firebase, I decided to move to Supabase and use this PostgreSQL database. It allowed me to create the structure for all the complex objects I wanted and needed and also create the vector and embeddings in the same tables. Additionally, I investigated the Datastore on Firebase, which I think has a similar approach, but my main concern with Firebase Datastore is the pricing. I think it is all implemented in cloud services, and I am not willing to pay that price, which is at least 10 pounds monthly. For now, I will stick with Supabase, but if that price changes in the future for Firebase, I would investigate it further.

The integration and setup were complex, but the pricing is decent. I am using the Supabase free trial, which allows two databases, but it would be nice to have more than those two. The biggest complexity lies in the knowledge required to get all these pieces together. The app I have implemented uses several technologies including KMP for building Android and iOS apps, Firebase, Supabase, n8n, webhooks, and everything built on GitHub. You have to have decent knowledge of all those technologies and how to handle the Google API console to create the APKs for the embeddings. The knowledge requirement is the most complex part, but apart from that, the pricing is not a significant issue.

I have not seen specific benefits in terms of time saved or improved efficiency yet because, as I said before, it was a proof of concept implemented in a basic app just to test it and be able to launch from the beginning to production. I was able to test it in production, and due to the good results and the positive feedback I got, I'm trying to increase the use of this technology in other apps and other parts of the software I have. I am definitely keen to keep exploring and making some research on how to implement this technology in other parts, and for sure, it was a great technology.

The most valuable part using Supabase Vector is how complex things I was able to build and deploy in a short period of time. Regarding the performance, I do not have enough data because, as I said, the app is still getting bigger and gaining more users. The number of users or the performance is not being a problem, or at least I did not have any issue in that regard. It is a technology I think is good and decent to be put in production. I am constantly looking for new problems or new features that I can implement and use this technology.

I have not seen a return on investment yet because I am not investigating those points. However, I can say that building these kinds of applications with so many complex components in less than a couple of months is amazing. The use of these technologies definitely impacts reducing the time and cost of implementation or deployment. It has a really good impact on companies, employees, and finances.

In this basic implementation or proof of concept project, I use the basic Supabase project available in the free trial. I am not sure which one of those options it falls under, but I use the free Supabase options to create two different databases—one for debugging or testing and the second one for release. I tend not to mix both environments. Apart from that, I use it with the full functions I needed using the free tier, which was amazing to be able to create all this technology for free, without the need to pay additional costs to use those technologies, apart from the embeddings ones from Google.

It was amazing to be able to create this software using Supabase Vector. I really recommend it and hope to see more improvements of this tool in the future.

My advice for others looking into using Supabase Vector is to take into account how you are going to create the embeddings, as it is not an option implemented straight away in Supabase. You will need to handle the embeddings with an external tool, so make sure to consider how these integrations are going to be made. I would rate this product an 8 out of 10.

My main use case for Supabase Vector is building a RAG pipeline, specifically for one of India's largest, one of the top five news channels, for their blogging website. We built a product where, on a news article, we show the relevant product according to that article and the user. We basically built a RAG pipeline based on that product, where it searches for the product according to the article and shows it to the user on the blog page.

For my use case, I wanted to show the relevant products according to the article to the user, and the setup was fairly straightforward. We extracted the details from the articles, including the category and other details, and used embedding to search the products' embedding that we had. We performed cosine similarities accordingly, leading to a simple and straightforward setup.

Supabase Vector specifically helped us match the relevant products to the articles and users through vector embedding, which has been a huge help. Supabase also has a lot of connections or trigger functions that you can use and manage from the dashboard itself, making it pretty helpful and useful.

The best features Supabase Vector offers include trigger functions and easy management from the dashboard. Triggering Postgres functions from the dashboard and enabling vector embedding, along with downloading those PG extensions, was super easy.

Out of those features, I find the entire dashboard to be the quickest part, along with using and testing the functions with the SQL editor, which is simple and straightforward.

Supabase Vector has positively impacted my organization by significantly reducing our testing time. Initially, when we used a traditional RDS database, our engineer struggled with proper testing, but once we moved to Supabase Vector, it got super easy with the dashboard and everything being clear. We shared credentials across giving access to our internal developers, allowing them to access and test cosine similarities of the vector, making the process simple compared to the complexity of connecting via terminal and SSH.

I estimate that we saved a lot of time after switching to Supabase Vector; it took us just two days to fully do the entire end-to-end testing setup. Initially, we planned it would take us more than two weeks if we had followed the normal route without using Supabase Vector, such as normal Postgres setups on a cloud server.

I think an improvement for Supabase Vector would be to have it enabled by default. Currently, you need to go through additional steps; we also deployed the open-source version of it, which lacks many features that the platform provides.

I would say the edge functions and logs of vectors and those triggers are often missing, especially in the open-source version of Supabase, which stands as a needed improvement.

Supabase Vector is fine as it is; I do not believe it needs any improvements. It is a perfectly fine product.

I have been using Supabase Vector for two to three years, and we were using it for a RAG pipeline.

Supabase Vector is stable; I find it to be pretty stable.

The scalability of Supabase Vector is impressive; it is pretty scalable and stable at the same time.

My experience with customer support is positive; I needed to reach out to Supabase, and the customer support team was pretty good, responding quickly to my issue. They resolved tickets efficiently, which I found very productive.

I would rate the customer support a nine since they replied quickly and answered my questions properly, which helped me a lot.

Before choosing Supabase Vector, we used AWS RDS as our previous solution, and we switched because we wanted a few of our engineers to test the RAG system and manipulate parameters with the SQL editor. We also needed to enable them to see the product data and its embedding, which was easier with Supabase Vector, making the migration from AWS pretty straightforward for our different use cases.

My experience with Supabase Vector's pricing, setup cost, and licensing is that it was pretty straightforward; it has an MIT license, which is good for us. We generally use a $200 to $300 subscription based on our database size, which is also fine for vectorization and other tasks.

We did not purchase Supabase Vector through the AWS Marketplace; instead, we simply did a git clone on an EC2 instance and ran it from there since Supabase is open-source software.

I have seen a clear return on investment; we saved money because RDS is pretty costly. Supabase includes all functions we were using, and we previously took backups manually, which was acceptable. The dashboard's management made access straightforward for users and super easy to maintain, resulting in very few errors. We completed what we initially estimated as a fifteen-day job in just three days, thanks to the Supabase dashboards and triggers.

We evaluated other options, specifically AWS RDS, and chose it based on its pricing and maintenance restrictions. We needed something that allowed proper developer access to the database without unnecessary security limitations, which Supabase Vector provides.

My advice for those considering using Supabase Vector is that they should definitely start with it; it is super easy and very easy to maintain. A backup setup with a Python script placed in an S3 bucket would also be good and save a lot of costs.

Supabase Vector is pretty good software that saves a lot of time and is super easy to use. I would rate this review a ten overall.

Our main use case for Supabase Vector is to use pgvector as the vector database solution to store our embeddings for large language model applications.

A specific example of how I'm using Supabase Vector for our large language model application is when we are building the Retrieval Augmented Generation (RAG) pipeline for an education application for one of our clients. We used the client's training material to perform chunking and embedding generation, and then stored the embeddings into Supabase Vector.

The major purpose of my use case with Supabase Vector is to store the embeddings.

I have been comparing many different similar products for storing embedded vectors, and I would say one advantage of Supabase Vector is that pgvector is embedded into Postgres, which makes it convenient for business use cases because we do not need to manage another data source.

Having pgvector embedded into Postgres helps our business use cases specifically as it is a very convenient way. We have Supabase basically as the host of most of our business relational database and user data, so since the client's applications are migrating to language model-empowered features, it is very useful, and we do not need to register for other database types.

Supabase Vector has positively impacted our organization as it is very convenient since our business databases are already hosted in Supabase, making integration easy. This convenience has led to measurable outcomes, as we also tried to use Pinecone and Weaviate, where we need to facilitate extra costs for storing the vectors. Now, since our business databases are already in Supabase, we only need to deal with one bill.

One improvement I feel Supabase Vector could benefit from is that Supabase SDK stands out when comparing with a conventional Postgres SDK, and it would be even nicer if we could have a more direct way for access.

I wish that there was a convenient way to make it compatible with the general Postgres database SDK, but this is actually not about the vector database; it is about the general Supabase SDK.

I have been using Supabase Vector for almost three years.

Supabase Vector is stable, and I have no complaints.

Supabase Vector's scalability works fine so far in our scale of applications, but I cannot tell what will happen if it scales out even more.

We usually just check the documentation or ask artificial intelligence for tech support, which is why we do not have much experience with customer support.

In my previous experience, we used Weaviate and Pinecone as the vector store, and the choice depends on the client's project requirement. For recent clients, the search was not dedicated for the Retrieval Augmented Generation pipeline for developers themselves, and we were using the Azure Search service, meaning we no longer directly used the low-level chunking and embedding technologies anymore.

I have seen a return on investment, as it obviously saves us a few hundred dollars every month compared with the approach of deploying the vector database on other providers.

I do not feel anything special about the pricing, setup cost, and licensing; we just regularly pay whatever we need, and I do not feel much difference.

Before choosing Supabase Vector, I evaluated Weaviate, Pinecone, and I also had a short experience with LanceDB, which is a local one similar to SQLite.

I do not have much to add about the features of Supabase Vector; I think so far this is good.

I would advise others looking into using Supabase Vector that if your business databases are already hosted on Supabase, pgvector on Supabase is definitely worth trying because it can save you a lot of time when you are bootstrapping and exploring this domain before making a decision to perhaps try using other solutions.

I rate this product an 8 out of 10.

I'm using Supabase Vector for the Postgres part. I use their Postgres database as the main requirement for the product from my side. If I am building a small website or any product, I don't need to create a backend for that part and host it. I can use Supabase APIs, such as their connection string, and it functions as a BaaS, backend as a service.

Supabase Vector is easy to set up and cost-effective because the alternative is Firebase, which requires a credit card. Most people don't want to use their credit card, so with Supabase Vector you can use the product without entering your card details. The services are straightforward and the database support is PostgreSQL, which is beneficial.

It's easy to set up and it has a free tier with PostgreSQL extensions with Supabase Vector. Every product I built has a free tier, so sometimes users can use the free tier. Some companies are still using the free trial because they didn't exceed their limit.

I think the support system can be better because after Supabase Vector stopped working in India, there is no support. Nobody knows how to deal with the database now. The naming structure is a little difficult. If you are naming a table, the guidance is sometimes not clear. It's a very simple feature to add sometimes but the process is very difficult. When I'm in Supabase Vector, there is a feature where I have to create a table. At the start, for newcomers, it's difficult, and then it becomes hard. You have to look for the documentation and YouTube videos to understand how it works.

Supabase Vector stopped working in India due to some government guidelines, so I have to switch with something else now.

I have been working with Supabase Vector for Postgres for around one to two years.

I would initially give Supabase Vector a seven for stability, but I would revise that to a five because there is currently downtime going on in India.

For the scalability part, I can give Supabase Vector an eight.

I recommend Supabase Vector to other users.

I previously used MongoDB instead of evaluating vector embeddings in Supabase Vector.

The setup for Supabase Vector is easy. I'm a developer, so I have set up the product myself. I have set up Supabase Vector by myself.

I purchase Supabase Vector directly from the vendor.

I compare Supabase Vector with Firebase. Supabase Vector is SQL and Firebase is NoSQL, so it's a bit different, but Firebase has more features than Supabase Vector.

I don't know about the efficiency of Supabase Vector in the case of high-dimensional data, and I currently don't have an idea of that part. It can be a bit cheaper. I would give this product an overall rating of eight.

My main use case for Supabase Vector is for the projects that I am building, as I use Supabase Vector to host the databases, and whenever I'm integrating AI into my application as well. Supabase Vector provides a vector database which is good for LLMs, and it provides good embedding as well.

One specific example of how I've used Supabase Vector in one of my projects is when I was working on a project where I had to build an RBAC matrix with different user roles, privileged user roles, and normal user roles. I used Supabase Vector to build these different tables and give different kinds of constraints on these tables. This ensures that only a normal user will be able to access the things that are applicable to them and also apply row-level security, which hardens the security of the system.

When I have to implement an LLM into my project or an AI chat base needs to be implemented, I use Supabase Vector for the databases as it is a good vector database to apply the natural embedding of the data.

The best features Supabase Vector offers include the ability to create a backend function or an API call, as I can use Supabase Vector's Edge Functions to implement those backend functionalities. Additionally, if I want to connect any OAuth provider like Google or LinkedIn, I can do that through Supabase Vector. It provides a seamless integration for that and it is really good if I want to integrate APIs, so it is a good vector database where I don't need to do much effort. Supabase Vector will do a lot of the running for me.

Out of the features I mentioned, I find myself using Edge Functions in Supabase Vector the most because I don't need a dedicated backend like Node.js. In this case, I can use the edge function and create an API call. I also find it very useful if I want to establish security in my project, which is very important; Supabase Vector does a lot of the running. It also flags any security vulnerabilities, so I can provide and harden my security using RLS or other similar features.

Supabase Vector has impacted my organization positively as it helps in strengthening the project without having dedicated backends, and it helps me establish a better workflow.

Regarding the positive impact, I believe Supabase Vector helps in reducing costs and fastening the process because a lot of it can be done easily. It is easy to establish the database as there is a clear step-by-step process on how I can connect my project to Supabase Vector. So it's really helpful and seamless that way.

I feel Supabase Vector can be improved by venturing further into adding more backend features and maybe working with fast API frameworks like Python languages; they could improve and widen their capabilities.

I think better role-based access can be provided to the users using Supabase Vector for further improvements.

I have been using Supabase Vector since three years ago, since I started working.

On a scale of one to ten, I would rate Supabase Vector as eight out of ten.

I chose eight out of ten because, as I said, it has a lot of capabilities and it is a very useful tool to use, so I would give a very strong rating of eight.

Regarding Supabase Vector's AI capabilities, I feel the security that Supabase Vector provides is really helpful and it really strengthens its capabilities. If one is talking about its support center and AI features used there, then it is good as well.

When it comes to Supabase Vector's AI capabilities, I find its output pretty accurate since I've been using Supabase Vector to integrate AI in my project. I find it pretty reliable and it is easy to integrate seamlessly into the software development life cycle as well.

I'm not sure how Supabase Vector was purchased; that is some other department's job, but this is how I use it.

I think I have mentioned everything that I wanted to mention.

My advice for others looking into using Supabase Vector is that the first learning curve might be a little challenging, but I would say spend as much time in the beginning as possible to learn all the things that Supabase Vector can do, how to connect your projects to Supabase Vector's API vectors, and learn everything in the beginning so you can seamlessly establish a process that works for you.

I feel Supabase Vector is doing a very great job in reaching out to companies and helping them. I believe every new entrepreneur who wants to build their own product, their own software product, should be using Supabase Vector, and a lot of them do as well. I think it's the best vector database out there. My overall rating for Supabase Vector is eight out of ten.

We have long PDF files we need to process, and then we need to ask questions and get answers on those files. My main usage is RAG for chats on PDFs, because models have limited memory in context, so less token usage and processing a large amount of data with question and answering capability.

I have been using Supabase Vector mostly for RAG. I have been using it for when a user is doing a lot of chats back and forth, and it is not useful to put all of the chat into the context of the LLM. I use it for chats on PDFs, and where I do not want to process and waste tokens.

In the same database I can utilize the embeddings and vectors provided by PG Vector. Additionally, Supabase Vector is a managed service, so I do not need to worry about scaling the database and managing the infrastructure. I can be up and running with Supabase Vector within a few seconds. The other valuable feature is semantic search, which I can do via PG Vector.

The managed service that Supabase Vector offers makes my workflow easier because I do not need to worry about managing the infrastructure or scaling the database. If I have an app idea and come up with a great app idea and I need to prototype and then push that idea into the market, I do not have to care about the infrastructure or scaling. In an instant, I get the managed Supabase Vector service available via the APIs, and then I just need to focus on my product.

The semantic search feature helps me specifically when working with my PDFs and chat use case because it is one of the core techniques behind modern AI chat systems and knowledge base assistants. Traditional keyword search may fail when words are different even though the meaning is nearly identical. For example, imagine a PRD that states the platform must support employee shift swapping with manager approval. When a user asks can workers exchange shifts, a keyword search may fail because exchange is not equal to swap and workers is not equal to employees. Instead of storing words, Supabase Vector stores embeddings, a vector.

An embedding is a list of numbers representing the meaning of text. When a user asks can workers exchange shifts, the system converts the question into the embedding and searches for nearby embeddings to find the PRD section about shift swapping and returns it to the LLM. The LLM now answers using the relevant section. If I have a 200-page PRD for my workforce management SaaS, chunking the document and generating the embeddings and then finding the closest chunk automatically gives me the meaning. It automatically searches for employees even though the query asks for workers. It really captures meaning instead of straightforward word-to-word search.

Supabase Vector positively impacts my organization by reducing the cost of the LLMs. Prior to using vector and RAG, I was giving whole chats to the LLM, and the LLM was processing the whole context and maintaining context. There is a limit in the LLMs of 1 million to 200,000 tokens I can process. With Supabase Vector, I have really reduced the cost and token consumption as well.

I do not have anything in my mind right now, but I can get back with this answer.

I do not have anything in my mind, but I will get back when I have more information.