Vocode Retrieval-augmented generation (RAG).


For Open Source developers building advanced conversational AI models, one challenge looms large: limited context windows in prompts. Past interactions quickly vanish, forcing bots to rely on incomplete information. Vocode introduces a novel solution: retrieval-augmented generation (RAG) powered by vector databases.

Here’s how it works:

  • Pre-computed embeddings: Vocode leverages vector databases like Pinecone to store concise representations of relevant knowledge. These bite-sized “memory chunks” efficiently encode factual details, contexts, and even emotional nuances.
  • Dynamic retrieval: As your bot interacts, Vocode queries the database for embeddings most similar to the current conversation. Think of it as a smart librarian fetching the most relevant books on the fly.
  • Contextual guidance: The retrieved embeddings are then fed to the language model, enriching its understanding of the ongoing dialogue. This enables the bot to generate more informed, contextually rich, and ultimately smarter responses.

Benefits for builders:

  • Scalability: Vector databases grow seamlessly with your needs, accommodating vast knowledge repositories without straining your system.
  • Efficiency: Say goodbye to bloated prompts filled with redundant information. Vocode delivers precise relevance with minimal data overhead.
  • Openness: Built on open-source principles, Vocode empowers developers to contribute, customize, and build upon its foundation.
  • Transparency: Explore the reasoning behind retrieved embeddings and gain valuable insights into your bot’s decision-making process.

How to set up your Pinecone database

In this guide, we’ll get your Pinecone database fired up and ready to fuel your retrieval-augmented chatbot with a robust knowledge base. Buckle up, developers!

1. Create Your Vector Index:

  1. Navigate to the Pinecone dashboard and grab the “Create Index” button.
  2. Name your index something meaningful, like “vocode-documentation-vector” for our Vocode-documentation bot.
  3. Set the dimensions to 1536 for compatibility with your chosen OpenAI model.
  4. Stick with cosine metric for consistency.
  5. Hit “Create Index” and let the magic happen!


Document loading script

You can manaully add documents to Pinecone any way you like as long as you include the required metadata. If you have a folder of PDFs, docx files, text files, etc. that you want to add to pinecone, you can use the below script which uses Unstructured to parse many kinds of files types, extract the text, and add it to pinecone.