GitHub Repos Translated

Chroma

For an AI app to answer from your own notes, it has to find the right ones first. Chroma is the search database programmers use to build that.

How you use it: A programmer installs it with one command, adds documents, and asks it to find the closest matches.

In plain English

The GitHub description calls Chroma search infrastructure for AI. The README shows a Python client you install with pip, a JavaScript client you install with npm, and a server you start with chroma run. The core demo is four operations: create a client, create a collection, add documents, and query the closest matches.

In the sample, Chroma handles tokenization, embedding, and indexing unless you pass your own embeddings. You can store metadata, such as which app a note came from, and filter on it later.

Chroma Cloud is the hosted service. The README says it covers vector, hybrid, and full-text search, and that a new database comes with $5 of free credits. The open-source project is Apache 2.0.

Practical ways to use it

  1. Example 1

    Prototype a document search on your laptop

    The README creates an in-memory client, adds two short documents, and queries the nearest matches to a question.

  2. Example 2

    Keep the index on disk

    The install blurb says client-server mode is chroma run with a folder path, so the data can outlive one Python process.

  3. Example 3

    Filter by where a note came from

    The sample stores metadata such as a source name, and the query snippet shows an optional where filter.

  4. Example 4

    Skip running a server

    Chroma Cloud is the hosted path in the README. The signup link is on trychroma.com, with $5 of credits stated in the README.

Who it is for

Developer. The happy path is a few lines of Python or JavaScript. You are embedding this inside an app, not opening a finished chat window.

Difficulty: Developer setup. The in-memory demo is short. Production means choosing persistence, embeddings, and who can reach the server.

Getting started

  1. 1

    Install the client

    The README shows pip install chromadb for Python and npm install chromadb for JavaScript.

  2. 2

    Run the four-call demo

    Create a client, create a collection, add documents with ids, and call query with a sentence. The README includes the snippet and a Colab notebook.

  3. 3

    Add a folder when you want the data to stick

    Use chroma run with a directory path, as the install comment describes, instead of only the in-memory client.

  4. 4

    Read the docs before you expose the port

    The README sends feature details to the Chroma docs. Do not put a database of private notes on an open network interface.

What you need first

Accounts
None for the local Python client. Chroma Cloud needs the signup linked in the README.
Hardware
A machine that can run Python or Node. The README does not list a minimum. Embeddings can be heavier than the database call itself.
Cost
The open-source project is Apache 2.0. Chroma Cloud includes $5 of free credits according to the README. Local use costs you the computer.

Cautions

  • An in-memory client forgets everything when the process stops. Use a path or the server mode if the notes matter.
  • Documents you add can be embedded by a model. Know whether that step calls a hosted embedding API.
  • The cloud credit is a starting balance, not a promise that a production app stays free.

Questions people ask

What does Chroma do?

It stores documents and finds the ones closest to a question, which is the search layer many AI apps use. You can run it locally or use Chroma Cloud.

Who is Chroma for?

Developers adding search to an app. The README's demo is Python code, with a JavaScript client as well.

What do I need first?

Python and pip install chromadb for the local demo. No account is required until you choose Chroma Cloud.

What should I watch out for?

The easiest client is in memory and disappears when the program exits. Persist to a folder before you load anything you need to keep.