Pinecone Vector Database - Annual Commit

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.

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.

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.

I have been using Pinecone for the last two years.

Pinecone is stable.

Pinecone is scalable.

I have not needed customer support yet, as everything works seamlessly.

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.

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.

The experience with pricing, setup cost, and licensing for Pinecone is not in my area, as I am a developer who uses the tools.

No other options were evaluated before choosing Pinecone.

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.

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.

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.

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.

I have been using Pinecone with code for one and a half years.

Pinecone is very stable.

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.

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.

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.

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.

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.

Before choosing Pinecone, I evaluated a few options, including Weaviate.

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