Cohere

Cohere's platform for chat text with streaming, embeddings, and reranking. Use it when you need a dedicated reranking backend that also serves chat and vectors, so one key covers the retrieve-then-rerank loop.

Capabilities

Text with streaming, embeddings, and reranking.

Configuration

Configure via environment variables. config/ai.php only forwards them with env() — set the values in .env or the host environment, never hardcode keys in the config file.

// In config/ai.php
return [
    'Ai' => [
        'providers' => [
            'cohere' => [
                'className' => 'Crustum\Ai\Providers\CohereProvider',
                'key' => env('COHERE_API_KEY'),
                'url' => env('COHERE_URL', 'https://api.cohere.com/v2'),
                'models' => [
                    'text' => [
                        'default' => env('COHERE_TEXT_MODEL', 'command-a-03-2025'),
                        'cheapest' => env('COHERE_TEXT_MODEL_CHEAPEST', 'command-r7b-12-2024'),
                        'smartest' => env('COHERE_TEXT_MODEL_SMARTEST', 'command-a-plus-05-2026'),
                    ],
                    'embeddings' => [
                        'default' => env('COHERE_EMBEDDINGS_MODEL', 'embed-v4.0'),
                        'dimensions' => (int)env('COHERE_EMBEDDINGS_DIMENSIONS', 1536),
                    ],
                    'reranking' => [
                        'default' => env('COHERE_RERANKING_MODEL', 'rerank-v4.0-pro'),
                    ],
                ],
            ],
        ],
    ],
];

Config key	Env var	Purpose
className	—	Provider class. Keep the default unless subclassing.
key (apiKey also accepted)	COHERE_API_KEY	Key from dashboard.cohere.com. Required.
url	COHERE_URL	API base URL. Change only for proxies or gateways.
models.text.default	COHERE_TEXT_MODEL	Chat model for normal calls.
models.text.cheapest	COHERE_TEXT_MODEL_CHEAPEST	Budget model.
models.text.smartest	COHERE_TEXT_MODEL_SMARTEST	Flagship model.
models.embeddings.default	COHERE_EMBEDDINGS_MODEL	Embeddings model.
models.embeddings.dimensions	COHERE_EMBEDDINGS_DIMENSIONS	Vector width. Must match the embeddings model.
models.reranking.default	COHERE_RERANKING_MODEL	Reranker used when no per-call model is given.
headers	— (config only)	Extra HTTP headers merged into every request. Add as a plain array if needed.

Model selection priority per call: explicit model: argument wins, then the matching models.* value above, then the provider's built-in fallback.

See also

• Providers — capability matrix
• Reranking — ordering search results
• Embeddings — vectors and caching