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¶

Option	Type	Default	Description
`url`	binary	`<<"http://localhost:11434">>`	Ollama server URL
`model`	binary	required	Model name

Supported Models¶

Popular embedding models available in Ollama:

Model	Dimensions	Notes
`nomic-embed-text`	768	Good general-purpose
`mxbai-embed-large`	1024	High quality
`all-minilm`	384	Lightweight
`snowflake-arctic-embed`	1024	Multilingual

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.