Run OpenClaw GDPR-compliant
OpenClaw + GDPR is feasible โ but only with the right configuration. Here's the path to a compliance-grade setup.
Model routing by data class
First rule: not everything goes to OpenAI/Anthropic. Sensitive data (personal, health, finance) routes to EU Mistral or local Llama via Ollama. Only non-sensitive workflows hit US providers โ and there with processor agreements.
Processor agreements (DPA)
Sign a DPA with each LLM provider: OpenAI offers an enterprise DPA, Anthropic via sales, Mistral via API plan. Hosting providers (Hetzner, AWS Frankfurt) too. We provide templates.
Data minimisation
Skills strip unnecessary personal data before LLM calls. Example: 'Manuel Streit, manuel@lol.marketing' becomes 'M. S., m@โฆ' โ pseudonymised but reconstructable in the OpenClaw daemon.
Retention policies
Logs, skill history, LLM calls with explicit deletion dates. Default: 90 days logs, 30 days detailed LLM calls, then deletion. Configurable in openclaw.json.
Subject rights
Access, rectification, erasure โ controllable via skill. /dsar export "name" generates a GDPR access report. /dsar delete "name" deletes all data about that person from logs and skills.