Cohere Provider on Bifrost

1) Chat Completions

Primary request path. Maps to upstream Cohere Chat API. Use extra_params for Cohere-specific fields.

curl -X POST http://localhost:8080/v1/chat/completions   -H "Content-Type: application/json"   -d '{
    "model": "cohere/command-r-plus",
    "messages": [{"role": "user", "content": "Hello"}],
    "top_k": 40,
    "safety_mode": "STRICT",
    "log_probs": true,
    "strict_tool_choice": false
  }'

2) Responses API

The Responses API uses the same upstream Cohere /v2/chat endpoint with conversion between OpenAI Responses format and Cohere's format. Gateway: POST /v1/responses. See Responses API in Bifrost docs.

Input & instructions

• String input converts to a user message; arrays convert to messages
• instructions becomes a system message (prepended to the message list)

Tool support

Supported tool type: function. Tool definitions and tool choice use the same conversions as Chat Completions.

Response conversion

• text → message | tool_use → function_call
• input_tokens / output_tokens preserved; cached tokens surfaced in token details when present

Streaming

• Event sequence: message-start → content-start → content-delta → content-end → message-end
• Tool call arguments accumulated across chunks; synthetic output_item.added events for text/reasoning
• Stable item IDs: msg_{messageID}_item_{outputIndex}

curl -X POST http://localhost:8080/v1/responses \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cohere/command-r-plus",
    "input": "Hello, how are you?",
    "top_k": 40,
    "stop": [".", "!"]
  }'

3) Embeddings

Embeddings via Bifrost at POST /v1/embeddings, routed to upstream Cohere /v2/embed. Non-streaming. See Embeddings in Bifrost docs.

Critical notes

• input_type is required for Cohere v3+ embedding models (Bifrost defaults to search_document when omitted)
• Use embedding_types to choose return formats (e.g. float, int8)
• Cohere-specific options (truncate, max_tokens) pass via the request body or extra_params in the Go SDK

Response conversion

• embeddings.float → data[].embedding
• meta.tokens → usage information
• Multiple embedding types are supported when requested

curl -X POST http://localhost:8080/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cohere/embed-english-v3.0",
    "input": ["text to embed"],
    "dimensions": 1024,
    "input_type": "search_query",
    "embedding_types": ["float"],
    "truncate": "START"
  }'

4) List Models

Lists available Cohere models through the Bifrost gateway. Upstream: GET /v1/models?page_size={defaultPageSize}. No request body required. See List Models in Bifrost docs.

• Model data is converted to the standard OpenAI-style models list format
• Pagination is cursor-based via next_page_token
• Optional filters endpoint and default_only are available via extra_params (Go SDK) or the request body (gateway)

curl http://localhost:8080/v1/models