Ollama

Chat text with streaming plus embeddings from a local Ollama daemon. Use it for offline development, eval loops, and air-gapped environments where no cloud call should happen. It points at http://localhost:11434 unless told otherwise, and the key may stay empty.

Capabilities

Text with streaming and embeddings.

Configuration

Configure via environment variables. config/ai.php only forwards them with env() — set the values in .env or the host environment.

// In config/ai.php
return [
    'Ai' => [
        'providers' => [
            'ollama' => [
                'className' => 'Crustum\Ai\Providers\OllamaProvider',
                'key' => env('OLLAMA_API_KEY', ''),
                'url' => env('OLLAMA_URL', 'http://localhost:11434'),
                'models' => [
                    'text' => [
                        'default' => env('OLLAMA_TEXT_MODEL', 'qwen3.5:4b'),
                        'cheapest' => env('OLLAMA_TEXT_MODEL_CHEAPEST', 'qwen3.5:0.8b'),
                        'smartest' => env('OLLAMA_TEXT_MODEL_SMARTEST', 'gemma4:cloud'),
                    ],
                    'embeddings' => [
                        'default' => env('OLLAMA_EMBEDDINGS_MODEL', 'nomic-embed-text'),
                        'dimensions' => (int)env('OLLAMA_EMBEDDINGS_DIMENSIONS', 768),
                    ],
                ],
            ],
        ],
    ],
];

Config key	Env var	Purpose
className	—	Provider class. Keep the default unless subclassing.
key (apiKey also accepted)	OLLAMA_API_KEY	Optional. Leave empty for a plain local daemon; set only when a reverse proxy in front of Ollama demands a token.
url	OLLAMA_URL	Ollama daemon base URL. Local default shown above; point at the LAN/VPN host for remote daemons.
models.text.default	OLLAMA_TEXT_MODEL	Ollama tag for normal chat calls (e.g. qwen3.5:4b).
models.text.cheapest	OLLAMA_TEXT_MODEL_CHEAPEST	Smallest/fastest local tag.
models.text.smartest	OLLAMA_TEXT_MODEL_SMARTEST	Most capable local (or proxied) tag.
models.embeddings.default	OLLAMA_EMBEDDINGS_MODEL	Embeddings tag.
models.embeddings.dimensions	OLLAMA_EMBEDDINGS_DIMENSIONS	Vector width of the embeddings tag. Must match the pulled model.
headers	— (config only)	Extra HTTP headers merged into every request (e.g. proxy auth). Add as a plain array if needed.

Model names are Ollama tags. Pull them with ollama pull <tag> before first use — unknown tags fail at request time, not at configuration time. Per-call model: overrides the configured tag.

Notes

The daemon must be reachable from the app host; remote Ollamas go in url, and reverse proxies that demand a token go in key. Pull a model before first use. Unknown tags fail at request time, not at configuration time.

See also

• Providers — capability matrix
• Embeddings — vectors and caching