Ollama Provider

Local embedding generation using Ollama server.

Requirements

• Ollama installed and running

Setup

# Install Ollama
brew install ollama  # macOS
# or download from https://ollama.ai

# Start server
ollama serve

# Pull an embedding model
ollama pull nomic-embed-text

Configuration

{ok, State} = barrel_embed:init(#{
    embedder => {ollama, #{
        url => <<"http://localhost:11434">>,  % default
        model => <<"nomic-embed-text">>       % required
    }}
}).

Options

Supported Models

Popular embedding models available in Ollama:

List available models:

ollama list

Example

%% Start application
application:ensure_all_started(barrel_embed).

%% Initialize
{ok, State} = barrel_embed:init(#{
    embedder => {ollama, #{
        url => <<"http://localhost:11434">>,
        model => <<"nomic-embed-text">>
    }}
}).

%% Single embedding
{ok, Vec} = barrel_embed:embed(<<"Hello world">>, State).
768 = length(Vec).

%% Batch embedding
{ok, Vecs} = barrel_embed:embed_batch([
    <<"First document">>,
    <<"Second document">>,
    <<"Third document">>
], State).
3 = length(Vecs).

API Compatibility

The Ollama provider supports both API versions:

• /api/embed (newer, preferred)
• /api/embeddings (legacy fallback)

It automatically detects and uses the appropriate endpoint.

Troubleshooting

Connection refused

Ensure Ollama is running:

ollama serve

Model not found

Pull the model first:

ollama pull nomic-embed-text

Slow first request

The first request may be slow as Ollama loads the model into memory. Subsequent requests will be fast.