Migrate from Anthropic

To migrate from the Anthropic API to Inworld Inference, you mainly change the client configuration. Standard message, streaming, and tool-use code keeps working through the Anthropic Messages API. Some Anthropic-specific parameters and content blocks are dropped, including top_k, PDF document blocks, and thinking blocks in history. Check supported and unsupported features before you migrate.

Key Differences

Migration Steps

1. Update Base URL and API Key

Python:

import os
import anthropic

api_key = os.environ["INWORLD_API_KEY"]  # Base64 credential copied from Portal

client = anthropic.Anthropic(
    # From:
    # (default: https://api.anthropic.com)
    # api_key="sk-ant-..."
    # To:
    base_url="https://api.inworld.ai",
    api_key=None,
    auth_token=api_key,
    default_headers={"Authorization": f"Basic {api_key}", "X-Api-Key": anthropic.omit},
)

# Standard requests stay the same
message = client.messages.create(
    model="anthropic/claude-opus-4-6",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Hello!"}
    ]
)

TypeScript:

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

const apiKey = process.env.INWORLD_API_KEY; // Base64 credential copied from Portal

const client = new Anthropic({
  // From:
  // (default: https://api.anthropic.com)
  // apiKey: 'sk-ant-...'
  // To:
  baseURL: 'https://api.inworld.ai',
  apiKey: null,
  authToken: apiKey,
  defaultHeaders: { Authorization: `Basic ${apiKey}` },
});

See Anthropic compatibility for why both settings are needed.

cURL:

# From:
# curl https://api.anthropic.com/v1/messages -H "x-api-key: sk-ant-..."
# To:
curl --request POST \
  --url https://api.inworld.ai/v1/messages \
  --header "Authorization: Basic $INWORLD_API_KEY" \
  --header 'anthropic-version: 2023-06-01' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "anthropic/claude-opus-4-6",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

2. Update Model Names (Optional)

Bare Claude model names work as-is: Inworld Inference finds the providers that serve the model and routes to one of them, with fallback between them. Add the anthropic/ prefix to pin a specific provider:

• claude-opus-4-6 → anthropic/claude-opus-4-6
• claude-opus-4-7 → anthropic/claude-opus-4-7
• claude-haiku-4-5-20251001 → anthropic/claude-haiku-4-5-20251001

3. Enable Intelligent Routing (Optional)

Instead of hardcoding a model, use Inworld Inference's intelligent routing:

message = client.messages.create(
    model="auto",  # Inworld Inference selects the best model
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Write a Python function to sort a list."}
    ]
)

What Stays the Same

• Message format: messages array with role and content, including text and image blocks.
• Streaming: stream=True works with Anthropic SSE events (message_start, content_block_delta, etc.).
• System prompt: Top-level system parameter (string or text blocks).
• Tool use: Tool definitions, tool_use content blocks, and tool_result messages.
• max_tokens, temperature, top_p, stop_sequences: Passed through.
• Multi-turn conversations: Same format.

What Changes

• Dropped fields: top_k, service_tier, mcp_servers, container, and document, search_result, and thinking blocks are silently ignored. Anthropic server tools are not executed.
• Response details: stop_sequence is always null, and thinking blocks have no signature.
• Errors: Router error bodies use the {"error": {"message", "type"}} shape rather than Anthropic's {"type": "error", ...} envelope. Authentication errors have a plain-text body.
• API key header: x-api-key is not read. Use the Basic authorization header shown above.

See supported and unsupported features for the full list.

What You Gain

1. Multi-provider fallbacks: List fallback models (or use a router). If Anthropic is down or rate-limited, Inworld Inference retries on OpenAI, Google, or other providers. The response still uses the Anthropic format.
2. Cost Optimization: Route simple queries to cheaper models, reserve Claude for complex tasks.
3. Cross-Provider Access: Use openai/gpt-5.4 or google-ai-studio/gemini-2.5-flash through the same Anthropic SDK — Inworld Inference translates the request and returns an Anthropic-format response.
4. Unified Observability: See all your LLM requests — across Anthropic, OpenAI, Google — in one dashboard.
5. Custom Routing: Build routers with conditional logic, A/B testing, and weighted variants.

Environment Variables

If you use environment variables, update them:

# From:
# export ANTHROPIC_API_KEY=sk-ant-...

# To:
export INWORLD_API_KEY=YOUR_INWORLD_API_KEY

Then pass INWORLD_API_KEY explicitly as shown in the client examples above. Don't rely on ANTHROPIC_API_KEY: the SDK sends it as x-api-key, which Inworld Inference does not read. For Claude Code, follow the Claude Code guide.

Support

Need help with a complex migration? Contact our engineering team.