embeviz

A simple webapp for visualizing text embeddings.

[!WARNING] THIS PROJECT IS WILDLY EXPERIMENTAL AT THE MOMENT! USE AT YOUR OWN RISK!

Click to view the capabilities:

The app consists of two components:

• A JSON API (written in Go, using gofiber framework)
• An SPA (written in JavaScript using ReactJS framework with React Router)

The SPA is available on the /ui URL path when running the app. The app also exposes the API docs on /api/v1/docs URL.

[!IMPORTANT] Before you attempt to access the SPA you must build it first! See the README for more details.

The swagger docs for the API are availble on /api/v1/docs.

The app leverages the go-embeddings Go module for fetching embeddings from various API providers like OpenAI, and chunking the input data if requested.

As a result of this you MUST supply API keys to the embeddings providers via environment variables when you run the app. See the README of the go-embeddings module for more details.

[!WARNING] By default the API stores the embeddings in an in-memory store (which is a major Go maps hack!). The only vector store currently supported is qdrant. See the docs.

Build

Build the UI:

cd ui && npm install && npm run build

[!IMPORTANT] You must build the UI app first as the Go binary embeds it as a static asset.

Build the Go binary:

go get ./... && go build

Nix

The project also provides a simple nix flake that leverages gomod2nix for consistent Go environments and builds.

To get started just run

nix develop

And you'll be dropped into a development shell.

[!NOTE] gomod2nix vendors dependencies into the local nix store so every time you add a new dependency you must run gomod2nix generate which updates the gomod2nix.toml file based on your go.mod/go.sum.

Run

[!IMPORTANT] Before you run the app you need to make sure you have set some environment variables required by specific AI embeddings providers. See the list below for the currently supported providers:

The project relies on the go-embeddings Go module so we only support specific AI embeddings API providers:

• OpenAI: OPENAI_API_KEY
• Cohere: COHERE_API_KEY
• Google VertexAI: VERTEXAI_TOKEN (get it by running gcloud auth print-access-token once you've set up your GCP project and authenticated locally) and GOOGLE_PROJECT_ID (the ID of the GCP project)

[!NOTE] If none of the above environment vars has been set, no AI embeddings provider is loaded and you won't be able to interact with the app. The project doesn't allow adding new embeddings providers from the UI app at the moment.

Once you've bundled the webapp and built the Go binary you can run the following command:

OPENAI_API_KEY="sk-XXXX" COHERE_API_KEY="XXX" ./embeviz

Alternatively you can also run the following commands, which installs node dependencies and builds the app on the fly:

npm --prefix ui install \
     && go generate ./...  \
     && OPENAI_API_KEY="sk-XXXX" COHERE_API_KEY="XXX" go run ./...

You should now be able to access the SPA on http://localhost:5050/ui.

The API docs should be available on http://localhost:5050/api/v1/docs.

TODO

• Clean up the code: both Go and especially React/CSS
• Add a text chunk hihglighter