remote/cohere¶

Uni-Xervo support¶

Provider ID: remote/cohere
Feature flag: provider-cohere
Capabilities: embed, rerank, generate, embed_multimodal (Cohere Embed v4, new in 0.13.0)

Authentication¶

Default key env var:

CO_API_KEY

Uni-Xervo provider options¶

api_key_env (string)
input_type (string, embedding requests)
embedding_dimensions (integer, override embedding dimensions)

Authoritative Uni-Xervo option schema:

https://github.com/rustic-ai/uni-xervo/blob/main/crates/uni-xervo/schemas/provider-options/cohere.schema.json

Authoritative model and config docs¶

Model catalog: https://docs.cohere.com/v2/docs/models
Chat/generation request config: https://docs.cohere.com/v2/reference/chat
Embeddings request config: https://docs.cohere.com/v2/reference/embed
Rerank request config: https://docs.cohere.com/v2/reference/rerank

Uni-Xervo generation options exposed¶

max_tokens
temperature
top_p

Example catalog entry¶

{
  "alias": "embed/cohere",
  "task": "embed",
  "provider_id": "remote/cohere",
  "model_id": "embed-english-v3.0",
  "options": {
    "api_key_env": "CO_API_KEY",
    "input_type": "search_document"
  }
}

Multimodal embed (Cohere Embed v4)¶

Cohere Embed v4 takes a mixed inputs: [{ "content": [...] }] payload of text and image blocks. Uni-Xervo converts MultimodalBlock::Text and MultimodalBlock::Image (URL or raw bytes — bytes are base64-encoded into a data:image/... URL upstream). MultimodalBlock::Audio is rejected because Cohere Embed v4 does not accept audio inputs.

{
  "alias": "embed/cohere-mm",
  "task": "embed_multimodal",
  "provider_id": "remote/cohere",
  "model_id": "embed-v4.0",
  "options": {
    "api_key_env": "CO_API_KEY",
    "input_type": "search_document"
  }
}

supported_modalities() reports [Text, Image]. EmbedResult::usage is populated from meta.billed_units.input_tokens when Cohere returns it.