Google Vertex AI Provider¶

Embedding generation using Google Vertex AI's embedding API. Provides access to Google's embedding models including the Gecko family.

Requirements¶

Google Cloud project with Vertex AI API enabled
Authentication via access token or API key

Configuration¶

Option 1: Access Token (Recommended for development)¶

Get a token with gcloud auth print-access-token:

{ok, State} = barrel_embed:init(#{
    embedder => {vertex, #{
        access_token => <<"ya29...">>,           % or use env var
        project => <<"my-project">>,
        region => <<"us-central1">>,
        model => <<"text-embedding-004">>
    }}
}).

Option 2: API Key¶

{ok, State} = barrel_embed:init(#{
    embedder => {vertex, #{
        api_key => <<"...">>,                    % or use env var
        project => <<"my-project">>,
        region => <<"us-central1">>,
        model => <<"text-embedding-004">>
    }}
}).

Options¶

Option	Type	Default	Description
`access_token`	binary	`GOOGLE_ACCESS_TOKEN` env var	OAuth access token
`api_key`	binary	`GOOGLE_API_KEY` env var	API key
`project`	binary	`GOOGLE_CLOUD_PROJECT` env var	GCP project ID
`region`	binary	`<<"us-central1">>`	GCP region
`model`	binary	`<<"text-embedding-004">>`	Model name

Using Environment Variables¶

# Get access token
export GOOGLE_ACCESS_TOKEN=$(gcloud auth print-access-token)
export GOOGLE_CLOUD_PROJECT=my-project

{ok, State} = barrel_embed:init(#{
    embedder => {vertex, #{}}
}).

Supported Models¶

Model	Dimensions	Notes
`text-embedding-004`	768	Default, latest
`text-embedding-005`	768	Newest
`textembedding-gecko@001`	768	Legacy
`textembedding-gecko@003`	768	Legacy
`textembedding-gecko-multilingual@001`	768	Multilingual legacy
`text-multilingual-embedding-002`	768	Multilingual

Setup Steps¶

Enable Vertex AI API
Go to GCP Console
Enable "Vertex AI API"
Get Authentication

For development (access token):

gcloud auth login
gcloud auth print-access-token

For production (service account): - Create service account with Vertex AI User role - Download JSON key - Use GOOGLE_APPLICATION_CREDENTIALS

Note Project ID
```
gcloud config get-value project
```

Example¶

%% Initialize
{ok, State} = barrel_embed:init(#{
    embedder => {vertex, #{
        access_token => <<"ya29...">>,
        project => <<"my-project">>,
        region => <<"us-central1">>,
        model => <<"text-embedding-004">>
    }}
}).

%% Generate embeddings
{ok, Vec} = barrel_embed:embed(<<"Machine learning is fascinating">>, State).

%% Batch
{ok, Vecs} = barrel_embed:embed_batch([
    <<"Document about AI">>,
    <<"Document about databases">>,
    <<"Document about networking">>
], State).

Regions¶

Vertex AI is available in many regions. Common choices:

us-central1 (Iowa) - Default
us-east1 (South Carolina)
europe-west1 (Belgium)
asia-northeast1 (Tokyo)

Access Token Refresh¶

Access tokens expire after 1 hour. For long-running applications:

Use service account authentication
Or refresh token periodically:

%% Refresh token
NewToken = os:cmd("gcloud auth print-access-token"),
{ok, NewState} = barrel_embed:init(#{
    embedder => {vertex, #{
        access_token => list_to_binary(string:trim(NewToken)),
        project => <<"my-project">>
    }}
}).

Why Vertex AI?¶

Google's latest embedding models
Tight integration with GCP services
BigQuery, Cloud Storage integration
Enterprise security (VPC-SC, CMEK)

Pricing¶

See cloud.google.com/vertex-ai/pricing for current rates.