Claude API with EU data residency
and honest trade-offs
OdinClaw routes to Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 via an EU-hosted, GDPR-compliant gateway. Same Anthropic pricing. No markup. Honest about the +20–40ms latency cost of the EU edge.
Claude models available on OdinClaw
All three Claude 4 tiers. 200K context across the board. Same pricing as Anthropic direct.
| Model | Model ID | Input / 1M | Output / 1M |
|---|---|---|---|
| Claude Opus 4.7 | claude-opus-4-7 | $5.00 | $25.00 |
| Claude Sonnet 4.6 | claude-sonnet-4-6 | $3.00 | $15.00 |
| Claude Haiku 4.5 | claude-haiku-4-5 | $0.80 | $4.00 |
Prices as of June 2026. See /models for the full registry with live pricing.
OdinClaw vs Anthropic direct — the honest comparison
We note where going direct is the better choice.
| Factor | OdinClaw | Anthropic direct |
|---|---|---|
| EU data residency | Yes — EU-hosted control plane | No — US-hosted (San Francisco / AWS us-east-1) |
| GDPR data processor | Yes — EU DPA available | Anthropic's standard DPA (US-entity) |
| Per-model pricing | At-cost passthrough — same as Anthropic | Direct pricing |
| Latency overhead | +20–40ms (EU edge hop) | 0ms overhead (direct) |
| Hard spend cap | Yes — request refused at limit | Soft cap / usage limits per tier |
| Free tier | 100K tokens/month, no card | No free tier — pay from first token |
| Multi-model access | Claude + DeepSeek + Gemini + Llama + more | Anthropic models only |
Honest latency note
Routing through OdinClaw adds approximately 20–40ms to each request — the cost of the EU-edge hop between our control plane and Anthropic's inference servers. For most workloads this is imperceptible. For latency-critical real-time applications, benchmark this before choosing the EU path. If Anthropic's direct latency is your hard requirement, go direct.
One-line migration from the Anthropic API
OdinClaw uses the same OpenAI-compatible API shape. The Anthropic Python SDK and openai-node both work — just change the base URL and key.
- OpenAI Python SDK (recommended)
- Anthropic Python SDK (with base_url override)
- openai-node / TypeScript
- Direct HTTP / curl
# Before (Anthropic direct — US-hosted)
from openai import OpenAI
client = OpenAI(
api_key=os.environ["ANTHROPIC_API_KEY"],
base_url="https://api.anthropic.com/v1",
)
# After (OdinClaw — EU-hosted, GDPR-compliant)
from openai import OpenAI
client = OpenAI(
api_key=os.environ["ODINCLAW_API_KEY"],
base_url="https://api.claw.odin-labs.ai/v1",
)
# Request body is identical — same model IDs
response = client.chat.completions.create(
model="claude-sonnet-4-6",
messages=[{"role": "user", "content": "Hello"}],
)What about my data?
OdinClaw acts as a data processor under GDPR. Your prompts are forwarded to Anthropic for inference and are not retained by OdinClaw beyond the network hop. We log billing metadata only: token counts, latency, model name, tier, and timestamp. No prompt or completion text is stored.
We are an early product. We do not currently hold SOC 2 or ISO 27001 certification. If your procurement process requires a certified vendor today, we will not pretend otherwise. Our security practices are documented in our privacy policy. We publish a changelog when our posture changes.
Common questions
Why use OdinClaw for the Anthropic API instead of Anthropic directly?
For EU-regulated businesses, the primary reason is data residency. Anthropic's API is hosted in the US. When your EU application calls api.anthropic.com directly, prompts leave the EU. If your prompts contain personal data — even pseudonymised data — this creates a cross-border transfer under GDPR that requires a valid transfer mechanism (SCCs, adequacy decision, or binding corporate rules). OdinClaw acts as an EU data processor: your request hits our EU-hosted control plane, we forward it to Anthropic for inference, and the response returns. Your legal relationship is with an EU-governed service. The trade-off is +20–40ms latency for the EU hop — a real cost worth knowing up front.
What is the latency penalty for routing through OdinClaw?
Approximately +20–40ms on top of Anthropic's base latency. This comes from the EU-edge hop between OdinClaw's control plane and Anthropic's inference servers. For most use cases — chatbots, document analysis, code generation, summarisation — this is imperceptible. For latency-critical applications (real-time voice, sub-100ms interactive UIs) the penalty matters and you should benchmark. If direct Anthropic latency is your ceiling, go direct.
Is the pricing the same as calling Anthropic directly?
Yes — OdinClaw passes through Anthropic model pricing at cost with no markup. Claude Sonnet 4.6 is $3.00/1M input and $15.00/1M output whether you call Anthropic directly or via OdinClaw. The EU advantage is additive, not a premium.
Which Claude models are available on OdinClaw?
Claude Opus 4.7, Claude Sonnet 4.6, and Claude Haiku 4.5. All three have 200K context windows. Model IDs are: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5. See /models for the full registry and any additions.
Does OdinClaw support Claude's tool-use and structured output?
Yes. The full /v1/chat/completions request shape is supported including tools, tool_choice, response_format (JSON mode), and streaming. Claude's native tool-use format works unchanged through our gateway.
Does OdinClaw hold SOC 2 or ISO 27001?
Not yet. We run on hardened EU infrastructure with TLS 1.3 in transit and least-privilege access controls. We are honest: if your procurement requires a certified vendor today, we are not there. Certification is on the roadmap and the changelog will reflect it when it ships.
Use Claude in the EU — GDPR-compliant
Free tier — 100K tokens per month, no credit card, hard spend cap at €0. Claude Opus 4.7, Sonnet 4.6, Haiku 4.5. EU data residency. Honest about +20–40ms.