Claude Code

Last updated:

On this page

Run Claude Code, the Anthropic Python SDK, and other Anthropic-compatible tools directly against staik — Swedish GPU infrastructure with full data residency. Under the hood this uses POST /v1/messages.

Quickstart

Point Claude Code at staik with two environment variables:

bash
export ANTHROPIC_BASE_URL=https://api.staik.se
export ANTHROPIC_API_KEY=sk-st-your-key

# Launch Claude Code
claude

Claude Code now uses staik's Swedish GPUs instead of Anthropic's cloud. All tools work — file reading, searching, code editing, and more.

Make it permanent

Add the variables to your .bashrc, .zshrc, or .env file so they are set automatically every time you open a terminal.

Important about token usage

Claude Code is token-intensive: each request sends the system prompt, tool definitions, CLAUDE.md files, and the full conversation history — often 20,000–30,000 prompt tokens per call. For serious usage we recommend one of the Agent or Coder plans (rolling hourly window) — see Plans and quotas. Prompt caching on /v1/messages makes recurring context cheaper automatically.

Anthropic SDK

Works the same way — set base_url:

python
import anthropic

client = anthropic.Anthropic(
    base_url="https://api.staik.se",
    api_key="sk-st-your-key",
)

message = client.messages.create(
    model="qwen3.6:35b-a3b",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain how a transformer works."}],
)
print(message.content[0].text)

Models and features

  • Send staik model names or Claude model names — Claude names are mapped automatically
  • Streaming, tool calling, system prompt, and prompt caching are fully supported
  • Same quotas, queues, and circuit breaker as the OpenAI endpoint

Limitations

See the Messages reference: extended thinking is ignored, images and PDF blocks are not supported via this endpoint — use chat completions for vision.