Migrate from Cohere to Railwail

TL;DR — Switch in Under 10 Minutes

• Drop the cohere-ai SDK — use the OpenAI SDK with Cohere models
• Command R+, Command R 08-2024 mapped to /v1/chat/completions
• Embed v4 mapped to /v1/embeddings
• Rerank v3.5 mapped to /v1/rerank (Cohere-compatible body)
• EU hosting, EUR billing, plus 274 other models

Why Move Off Cohere?

Cohere has strong enterprise-grade embeddings, reranking, and Command R+ for RAG workloads. The trade-offs are limited frontier-model variety, US-default hosting (EU is enterprise-tier only), and a proprietary SDK that does not play well with OpenAI-shaped codebases. Railwail mirrors Cohere's flagship models behind the OpenAI Chat Completions / Embeddings interface, adds Cohere-compatible /v1/rerank, and unifies billing in EUR.

Step 1 — Get a Railwail API Key

Sign up at railwail.com and generate a key.

Step 2 — Replace the Cohere SDK

Chat — TypeScript

Before (Cohere):

import { CohereClient } from "cohere-ai";

const cohere = new CohereClient({ token: process.env.COHERE_API_KEY });

const res = await cohere.chat({
  model: "command-r-plus-08-2024",
  message: "Hello",
});

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.RAILWAIL_API_KEY,
  baseURL: "https://api.railwail.com/v1",
});

const res = await client.chat.completions.create({
  model: "command-r-plus-08-2024",
  messages: [{ role: "user", content: "Hello" }],
});

Embed — Python

from openai import OpenAI
client = OpenAI(
    api_key=os.environ["RAILWAIL_API_KEY"],
    base_url="https://api.railwail.com/v1",
)

resp = client.embeddings.create(
    model="embed-english-v4",
    input=["hello world", "another doc"],
)
print(resp.data[0].embedding[:5])

Rerank — cURL

curl https://api.railwail.com/v1/rerank \
  -H "Authorization: Bearer $RAILWAIL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "rerank-v3.5",
    "query": "What is the capital of France?",
    "documents": ["Paris is the capital of France.", "Berlin is in Germany."],
    "top_n": 1
  }'

API Endpoint Mapping

Model Mapping

Pricing Comparison (per 1M tokens, May 2026)

Why Railwail Over Cohere

• EU billing in EUR with VAT receipts
• OpenAI-compatible API — drop the cohere-ai SDK
• Same Cohere models at same price
• Plus Claude, GPT-4o, Gemini, Voyage embeddings for cross-provider RAG experiments
• Built-in /v1/rerank endpoint that accepts Cohere's request body
• Built-in playground for embed-quality A/B testing

FAQ

Is the rerank API exactly Cohere-compatible?

Yes. Same query, documents, top_n, return_documents fields. Response includes results array with index and relevance_score.

What about Cohere's RAG with citations?

Cohere's stateful chat (conversation_id, connectors, documents in chat call) is proprietary. Railwail supports the stateless OpenAI Chat Completions schema — pass retrieved documents in the system or user message yourself. Citations metadata is preserved in the response when using Cohere models.

Are embedding input types (search_query vs search_document) preserved?

Yes. Pass input_type: 'search_query' or 'search_document' as an extension parameter.

Can I do binary / int8 embeddings?

Yes. Pass embedding_types: ['binary'] or ['int8'] to get quantised embeddings for vector-store storage efficiency.

Will my Cohere tool-use prompts work?

Cohere's tools format is converted to OpenAI tools schema by Railwail. Tool calls in the response come back as tool_calls per OpenAI convention.

Next Steps

• Sign up at railwail.com
• Generate an API key
• Drop cohere-ai and use the OpenAI SDK
• Use /v1/rerank for reranking workloads
• Read the reference at railwail.com/docs
• Browse Cohere models at railwail.com/models?provider=cohere
• Compare pricing at railwail.com/pricing