Cohere¶

Setup¶

pip install llmgate[cohere]
export COHERE_API_KEY="..."

Model prefix¶

Always prefix with cohere/:

completion("cohere/command-r-plus", messages)
completion("cohere/command-r", messages)
completion("cohere/command-a-03-2025", messages)

Embeddings¶

Cohere has excellent embedding models with task-aware vectors.

Always set input_type

The default (search_document) is wrong for query embedding. Use the correct type for production RAG.

from llmgate import embed

# Embedding corpus documents:
resp = embed("cohere/embed-english-v3.0", chunks,
             input_type="search_document")

# Embedding a search query:
resp = embed("cohere/embed-english-v3.0", query,
             input_type="search_query")

# Multilingual:
resp = embed("cohere/embed-multilingual-v3.0", texts,
             input_type="search_document")

# Truncation strategy for long inputs:
resp = embed("cohere/embed-english-v3.0", long_text,
             input_type="search_document",
             truncate="END")   # "NONE" (error) | "START" | "END"

Supported input_type values: "search_document", "search_query", "classification", "clustering".

Vision¶

Not supported

Cohere's vision API is not yet stable. Passing image content to a Cohere model raises VisionNotSupported.

from llmgate.exceptions import VisionNotSupported

try:
    resp = completion("cohere/command-r-plus", messages_with_images)
except VisionNotSupported:
    resp = completion("gpt-4o-mini", messages_with_images)  # fallback