GitHub

Chroma: The Open-Source Embedding Database

What is the project about?

Chroma is an open-source database specifically designed for storing and searching embeddings. Embeddings are numerical representations of data (text, images, audio, etc.) that capture their semantic meaning. Chroma makes it easy to build applications that leverage the power of embeddings, particularly in conjunction with Large Language Models (LLMs).

What problem does it solve?

Traditional databases are optimized for searching based on exact matches or substrings. Chroma addresses the need to search based on semantic similarity. This is crucial for tasks like:

  • Finding documents that are conceptually related, even if they don't share the same keywords.
  • Building recommendation systems based on the underlying meaning of items.
  • Providing contextually relevant information to LLMs, enabling them to "remember" and reason about data. It essentially gives LLMs a form of long-term memory.

What are the features of the project?

  • Simple API: Easy to use with a minimal, 4-function core API for adding, querying, updating, and deleting data.
  • Integrations: Works seamlessly with popular LLM frameworks like LangChain (Python and JavaScript) and LlamaIndex.
  • Scalability: Designed to work in development, testing, and production environments, from local notebooks to large clusters.
  • Feature-Rich: Supports queries, metadata filtering, and density estimation.
  • Open Source: Freely available and licensed under Apache 2.0.
  • Automatic Embedding: Chroma can handle the tokenization, embedding, and indexing of documents automatically, or you can provide your own embeddings.
  • Client-Server Mode: Supports running as a client-server setup for more robust deployments.

What are the technologies used in the project?

  • Python: Primary language for the core database and client.
  • JavaScript: Client library for JavaScript/Node.js applications.
  • Embedding Models:
    • Sentence Transformers (default: all-minilm-l6-v2)
    • OpenAI Embeddings
    • Cohere Embeddings
    • Support for custom embedding functions.
  • pip and npm for package management.

What are the benefits of the project?

  • Fast Development: Quickly build LLM applications with memory capabilities.
  • Easy to Use: Simple API reduces the complexity of working with embeddings.
  • Scalable: Grows with your application's needs.
  • Cost-Effective: Open-source and free to use.
  • Community Support: Active Discord community and open development process.

What are the use cases of the project?

  • "Chat your data" applications: Build chatbots that can answer questions based on a corpus of documents.
  • Semantic Search: Implement search engines that understand the meaning of queries.
  • Recommendation Systems: Recommend items based on semantic similarity.
  • Question Answering: Create systems that can answer questions based on a knowledge base.
  • Document Summarization: Use LLMs with Chroma to summarize large documents.
  • Code Search: Search codebases based on functionality rather than just keywords.
  • Any application that benefits from understanding the meaning of data. This includes image search, audio search, and more.
</p>
chroma screenshot