Migrate from Anthropic Console to Railwail

TL;DR — Switch in Under 5 Minutes

• Railwail supports BOTH the Anthropic Messages API (/v1/messages) AND the OpenAI Chat Completions API (/v1/chat/completions) for Claude models
• Keep @anthropic-ai/sdk: change base_url to https://api.railwail.com/anthropic and that is it
• Or use the OpenAI SDK with model='claude-sonnet-4-6' — one less dependency
• All Claude 4 features supported: long context, tool use, computer use, prompt caching, vision
• EU-hosted endpoint, EUR billing, plus access to 274 other models behind the same key

Why Move Off Anthropic Console Direct?

Anthropic's Claude is the gold standard for long-context reasoning, agentic workflows, and instruction following. But the Anthropic Console is USD-only, US-hosted, and requires a separate SDK from your OpenAI integration. If your stack already speaks OpenAI Chat Completions and you want EUR billing, Railwail lets you keep using Claude with zero code refactor — or with even less code than before.

Teams typically migrate when they (a) need EUR invoices for VAT compliance, (b) want to unify Claude and GPT-4o calls under a single SDK, (c) require EU data residency, or (d) hit Anthropic's tier-based rate limits and want a higher default ceiling.

Step 1 — Get a Railwail API Key

Sign up at railwail.com, navigate to Dashboard → API Keys, click Create New Key. Free credits are included so you can validate the migration without committing to a paid plan.

Step 2 — Choose Your SDK Path

Railwail supports two equally-valid migration paths. Pick whichever requires less code change.

Path A — Keep the Anthropic SDK

If your code uses @anthropic-ai/sdk and relies on Anthropic-specific request shapes (Messages API with system as a top-level field, multi-block content arrays, tool_use blocks), this is the lowest-friction path. Change one line.

Before (Anthropic direct, TypeScript):

import Anthropic from "@anthropic-ai/sdk";

const anthropic = new Anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY,
});

const msg = await anthropic.messages.create({
  model: "claude-sonnet-4-5",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello Claude" }],
});

import Anthropic from "@anthropic-ai/sdk";

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

const msg = await anthropic.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello Claude" }],
});

Path A — Python equivalent

from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["RAILWAIL_API_KEY"],
    base_url="https://api.railwail.com/anthropic",
)

msg = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello Claude"}],
)
print(msg.content[0].text)

Path B — Use the OpenAI SDK

If you want one SDK across all models, replace @anthropic-ai/sdk with the OpenAI client. Railwail accepts Claude model IDs in standard chat.completions calls and handles the schema translation internally.

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: "claude-sonnet-4-6",
  messages: [
    { role: "system", content: "You are concise." },
    { role: "user", content: "Hello Claude via OpenAI SDK" },
  ],
});
console.log(res.choices[0].message.content);

Path B — cURL

curl https://api.railwail.com/v1/chat/completions \
  -H "Authorization: Bearer $RAILWAIL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Hello Claude"}],
    "max_tokens": 1024
  }'

Step 3 — Verify Tool Use and Streaming

Tool use translates correctly across both paths. On Path A (Anthropic SDK), your tool definitions and tool_use response blocks work unchanged. On Path B (OpenAI SDK), Railwail converts between OpenAI's tools/tool_calls schema and Anthropic's tool_use blocks transparently. Streaming SSE events are passed through in the native format of whichever SDK you chose.

Step 4 — Prompt Caching

Anthropic's prompt caching feature (cache_control: ephemeral) is fully supported on Railwail Path A. Cache hits get the same 90% discount you would see on Anthropic Console direct. On Path B, prompt caching is automatically enabled when you set the cache_marker field on system or user messages — see railwail.com/docs/caching.

API Endpoint Mapping

Claude Model Mapping

Pricing Comparison (per 1M tokens, May 2026)

Why Railwail Over Anthropic Console

• EU billing in EUR with VAT receipts — clean accounting for European entities
• Higher default rate limits than Anthropic's tier-based system
• Two API surfaces — keep Anthropic SDK, or unify on OpenAI SDK
• Access to GPT-4o, Gemini, Mistral, Llama through the same key
• Built-in playground to A/B Claude vs other models on real prompts
• Prompt caching, tool use, computer use, vision — all preserved
• Spend caps and per-key isolation for multi-tenant apps

FAQ

Does prompt caching still get the 90% discount?

Yes. Anthropic's cache_control: ephemeral mechanism is forwarded directly. Cache writes cost 25% more than base input, cache hits cost 90% less. Pricing parity is preserved end-to-end.

What about computer use and the screenshot tool?

Computer use is fully supported on Path A (Anthropic SDK). The tool_use blocks for computer-screenshot and computer-action work unchanged. See the dedicated computer-use cookbook at railwail.com/docs/anthropic/computer-use.

Can I use the Messages API with vision (images)?

Yes. Image content blocks with base64 or URL sources work identically. There is no size limit beyond Anthropic's native 5MB-per-image cap.

What about Anthropic's Workbench and the saved prompts?

Workbench is a separate UI feature of Anthropic Console. Railwail's playground at railwail.com/models offers equivalent functionality — saved prompts, side-by-side model comparison, and one-click export to TypeScript/Python/cURL.

Do tool_choice and disable_parallel_tool_use parameters work?

Yes. The full Anthropic Messages parameter set — tool_choice (auto, any, tool), disable_parallel_tool_use, system blocks with cache_control, metadata.user_id, stop_sequences — is forwarded verbatim.

What happens if Anthropic upstream has an outage?

Railwail does not auto-fallback to a different model by default (that could silently change output quality). You can opt in to fallback rules in the dashboard, e.g. 'route Claude Sonnet → GPT-4o on 5xx'. Most teams prefer explicit error handling.

Next Steps

• Create your Railwail account at railwail.com
• Generate an API key in Dashboard → API Keys
• Pick Path A (Anthropic SDK) or Path B (OpenAI SDK) — both work
• Update one line: base_url
• Read the full Anthropic-compatible reference at railwail.com/docs/anthropic
• Browse the Claude models at railwail.com/models?provider=anthropic
• Compare per-token pricing at railwail.com/pricing