RAG workflows have transformed document research and now provide precise answers with citations
What is our primary use case?
My main use case for Pinecone is creating vector indexes for GenAI applications.
A specific example of how I use Pinecone in one of my projects is utilizing a RAG pipeline where I take text from PDF documents, convert those into chunks, ingest those into the Pinecone vector database, and then have a frontend UI that uses LLMs to query the vector database and retrieve answers.
What I appreciate about Pinecone is that it provides reranking and other features, and it's a SaaS-based solution that is serverless.
What is most valuable?
Pinecone's reranking aspect works by taking a list of documents from the indexes and organizing them based on the ranking that is relevant to the question being asked by the user, ensuring that if reranking is applied, the user gets the most relevant answers as LLMs understand them, providing near-perfect answers versus when not using reranking, where the LLM takes all output from the vector index, which won't be quite that perfect.
Pinecone's serverless aspect is valuable because I don't have to manage the infrastructure myself, as Pinecone takes care of that.
Pinecone has positively impacted my organization by helping people in needle-in-a-haystack situations, as previously they had to grind through PDF documents, PowerPoint documents, and websites, but now with Pinecone, they can ask questions and receive references to documents along with the page numbers where that information exists, so they can use it as a reference or backtrack, especially for things such as FDA approvals where they can quote the exact page number from PDF documents, eliminating hallucination and providing real-time data that relies on an external vector database with enough guardrails to ensure it won't provide information not in the vector database, confining it to the information present in the indexes.
Pinecone has helped full-time employees rely less on contractors to find information, enabling them to access data at their fingertips and reducing the turnaround time to generate reports.
What needs improvement?
I give Pinecone a nine out of ten because I hope it provides an end-to-end agentic solution, but currently, it doesn't have those agentic capabilities, meaning I have to create a Streamlit application and manage it to communicate with Pinecone. If Pinecone could provide those kinds of web apps out of the box, I would give it a perfect ten.
Nothing else is needed since Pinecone provides APIs for integration, making it not a hurdle, and I am happy with what I have.
Pinecone is good as it is, but had it been on AWS infrastructure, we wouldn't experience some network lags because it's outside AWS. However, when we started two years ago, there weren't any vector databases on AWS, making Pinecone a pioneer in the field.
For how long have I used the solution?
I have been using Pinecone for the last two years.
What do I think about the stability of the solution?
What do I think about the scalability of the solution?
How are customer service and support?
I have not needed customer support yet, as everything works seamlessly.
How would you rate customer service and support?
Which solution did I use previously and why did I switch?
There was no solution before Pinecone, as the vector database gained traction about two years ago, and Pinecone were the pioneers in this field, which is why we picked them.
What was our ROI?
I have seen a return on investment with Pinecone, as the application we built received positive feedback from internal stakeholders about how much it's helping them make business decisions and access information quickly at their fingertips.
What's my experience with pricing, setup cost, and licensing?
The experience with pricing, setup cost, and licensing for Pinecone is not in my area, as I am a developer who uses the tools.
Which other solutions did I evaluate?
No other options were evaluated before choosing Pinecone.
What other advice do I have?
Pinecone perfectly fits my organization's needs based on our use case. The market for vector databases is broad right now, offering many options; however, I don't have experience with other tools and technologies. I would give Pinecone a rating of nine out of ten overall.
Faced challenges with metadata filtering but have achieved reliable long-term memory for chat applications
What is our primary use case?
My main use case for Pinecone involves storage of chat data, specifically chat transcripts, and retrieval of matched chat messages.
We store chat transcripts as vectors in Pinecone. When we have a new chat message, we utilize a retrieval mechanism to match and find the last five messages so that it can act as a memory. Essentially, Pinecone serves as a long-term memory for our application, while we use Redis for our short-term memory.
What is most valuable?
We were looking at multiple options for a vector database, and we found Pinecone to be the easiest to integrate into our solution. Plus, it has a very generous free tier, which helps us as a startup.
The best features Pinecone offers are quick setup and good indexing for us. The retrieval mechanisms are fast, and the integration with Python as with JavaScript and TypeScript libraries that Pinecone provides are very robust. Authentication is also very good.
The namespaces feature allows us to break down or store data for each user separately, reducing interference and maintaining privacy as an important feature.
Pinecone has positively impacted our organization by enhancing efficiency for the team, and the long-term effect has been that the chats have become much more personalized due to the memory added through a vector database. We are seeing that the trainees getting trained on the platform are more satisfied with the results or messages generated by AI.
What needs improvement?
One major issue I have noticed with Pinecone is that it does not allow me to search based on metadata. This can cause problems because while vector indexing or vector search is good, if you populate certain categories of messages or metadata into a vector database, searching through the data using the filter of metadata is not possible.
For our requirements, Pinecone is more than enough. If improvements are required, I would suggest taking a look at the embeddings and possibly improving the embedding sizes.
For how long have I used the solution?
I have been using Pinecone with code for one and a half years.
What do I think about the stability of the solution?
What do I think about the scalability of the solution?
Pinecone's scalability is pretty decent for us, as we have not encountered issues. We are storing close to around 600K items or entries in the database, and our indexing and retrievals are within seconds, often in microseconds.
How are customer service and support?
The customer support of Pinecone is very good; you send an email and receive a response within a few hours, typically four to five hours. Additionally, you can set up a call if needed.
Since we are on the minimal plan, I would rate the customer support around 8 out of 10.
How would you rate customer service and support?
Which solution did I use previously and why did I switch?
We previously tried setting up with Weaviate and another solution. During my research, we checked out a couple of options, including an on-prem solution that I tried to set up on my machine, but it was very painful, so we went with the cloud service provider because the setup was nearly nonexistent.
How was the initial setup?
The setup cost for us is nil, and the licensing and pricing are pretty decent. Pinecone works on the storage amount, so our bills are pretty low, and we are good.
What's my experience with pricing, setup cost, and licensing?
The setup cost for us is nil, and the licensing and pricing are pretty decent. Pinecone works on the storage amount, so our bills are pretty low, and we are good.
Which other solutions did I evaluate?
Before choosing Pinecone, I evaluated a few options, including Weaviate.
What other advice do I have?
I would suggest that Pinecone is one of the best options available. I would rank it in the top three for vector databases and qualify it as number one in the market. There are many others such as Weaviate and Milvus, but they come with certain issues such as lacking a free tier or having a very low one.
Moreover, solutions like Milvus and FAISS are on-prem, which makes setup and stability a pain, primarily catering to big enterprises. For startups, Pinecone is indeed the best.
We are just a client of Pinecone; we do not have any other business relationship.
Rating: 4/5
Nice vector db easy to use
What do you like best about the product?
its provide various of features and great vector db support
What do you dislike about the product?
may be it is close source and needed some features which are not there yet.
What problems is the product solving and how is that benefiting you?
The latency is very minimal and provide large search/retrieval with fully managed serverless infrastructure
ideal for machine learning, AI applications and similarity search
What do you like best about the product?
It is specialised in AI driven use cases with real time and low latency search giving seamless integration into machine learning workflows with scalable infrastruture optimized for unstructured and semi-structured data in AI applications.
What do you dislike about the product?
It has limited focus that is related only with the vector data with no major focus on Business intelligence in data transformation tool.
Also it's use case is little complex with lack of ecosystem integration.
What problems is the product solving and how is that benefiting you?
It is solving the issue related with AI vector data generated from the app.
God of creating embeddings
What do you like best about the product?
when iam creating embeddings,compared to other products,it feels hassle free& cheap.
What do you dislike about the product?
I am the beta tester of pinecone AI assiatant,it is not production ready so it feels like only for testing,i am expecting for the production ready version.
What problems is the product solving and how is that benefiting you?
hassle free functions and embeddings data sets
Pinecone assistant beta user
What do you like best about the product?
I have been using pinecone for embeddings and it is cheaper and reliable compared to other embedding services.
What do you dislike about the product?
I dislike the overall feel which feels lightweighed for the product service documentation.
I love to see pinecone assistant in deployable version because it is powerful yet it is in the beta version only for testing not for production
What problems is the product solving and how is that benefiting you?
Creating embeddings at ease without any big pricing.
Good support from team.
Solid option for vector DB
What do you like best about the product?
Easy to use. very reliable and fast. Competitive price
What do you dislike about the product?
Maybe some extra features would be nice, and some more clarity into its AKNN algo, which is hidden from the user
What problems is the product solving and how is that benefiting you?
Finding scientific documents in very large volumes of Data.
Pinecone: The Backbone of Efficient Vector Search and Retrieval
What do you like best about the product?
Pinecone excels in providing a seamless, high-performance vector search experience. Its ease of use, combined with powerful features like real-time updates and scalability, makes it a go-to solution for managing complex vector data. The ability to effortlessly integrate with existing workflows and its top-notch customer support are definite highlights.
What do you dislike about the product?
While Pinecone is robust, the pricing can be a bit steep for smaller projects or startups. Additionally, more granular control over indexing options would enhance customization for advanced users. However, the benefits far outweigh these minor drawbacks.
What problems is the product solving and how is that benefiting you?
Pinecone is solving the complex challenge of efficient and scalable vector search. In an era where managing large volumes of high-dimensional data is critical, Pinecone's ability to index, search, and retrieve vectors quickly and accurately is a game-changer. For us, this means faster query responses, enhanced data retrieval accuracy, and the ability to focus on building better products rather than managing infrastructure. Pinecone's solution has drastically reduced the time and effort required to manage and search vector data, allowing our team to be more productive and innovative.
Using Pinecone on production - 1 year later
What do you like best about the product?
Pinecone was our primary choice and we have not considered changing since.
- High performance (upsert and search in the ms)
- Simple integration via API and deployment and now after their recent release of serverless indexes it's very simple to maintain and scale (it's autoscaling).
- Low price (relative to the number of vectors) and free limited indexes. Free indexes are great to run development environment data. For a while it was impossible to upgrade a free index to a paying one, but this is now addressed.
- Incredible support (we had an issue and was not expecting getting this quality of support without paying the usual business support fees of an AWS for example)
- The ability to assign metadata is very useful (we still maintain a traditional db to keep track of the vectors)
- The single stage query vector/metadata is very useful and saves the headache of over-querying
- One feature we have meant to use is the use of sparse vectors in combination with the dense vectors. So, can't really comment yet
What do you dislike about the product?
Love most of it as is
- The documentation using metadata and single stage queries is a bit light
- They have a smart bot to help answer support questions. On the great side, it seems they use their own technology for RAG type of application, but on the other it often misses the mark. ChatGPT or Perplexity are surprisingly more effective.
- There has been a few down times, but they are very communicative about them and maintain a server health page for each endpoint. It's usually related to a specific infrastructure (AWS or GCP) they run on.
- They have been growing and improving the technology, and like with other player, sometimes to update their python library or the way to reference to the indexes. But each time it's been toward simplification, and I suspect it will stabilize.
What problems is the product solving and how is that benefiting you?
Semantic matching
A great serverless DBaaS for vectors
What do you like best about the product?
Pinecode offers a simple API and lean management interface for a completely low maintenance vector storage and query solution.
What do you dislike about the product?
I started using Pinecone when it was new and had some rough edges. But support was proactive and smart. In the last year I can say there is nothing to not like. It has been awesome.
What problems is the product solving and how is that benefiting you?
We use Pinecone's serverless platform (on AWS) for vector search. Our vector dimension is 3072. Part of our use is user queries. The performance has been excellent and scalability is automatic. We also use the query capability in other parts of our stack where performance is not so important but reliability is a factor.
