D3x vs building in-house: the demo is easy. Production is the hard part.

Scoped tool access so agents only call permitted PMS and ops actions; role-based access control and enterprise SSO; human-in-the-loop and escalation rules; confidence-aware handoff; full audit logs with the agent's rationale for every decision; EU hosting and GDPR alignment; and a hard rule that customer data is never used to train third-party foundation models. Building the same control plane yourself is a product in its own right.

Codified hospitality logic above the model: hundreds of PMS and OTA quirks (for example, Booking.com reservation IDs gaining a "_1" suffix through the channel manager), policy hooks, escalation rules, and decision logging. This is what turns a capable model into 60–70% autonomous resolution in production, without learning every edge case as a guest-facing incident.

No. Every D3x customer gets an MCP server, so your engineers can control and extend D3x programmatically from Claude or any LLM. You get the customization surface of an internal platform without owning connector maintenance, model upgrades, channel certifications, and the security control plane.

In-house means permanent engineering, security, and compliance headcount plus inference and channel costs, with every LLM API change triggering policy re-testing. D3x starts at €5 per room per month with public SMB pricing and includes continuous LLM upgrades, the learning loop, and the AI knowledge base, with enterprise plans scoped on properties, channels, and volume.

Not automatically. In-house only wins if you fund continuous audits, retention/erasure workflows, penetration testing, and AI-specific threat models. D3x is EU-hosted and GDPR-aligned, with SOC 2 Type II in progress, RBAC, SSO for enterprise, and per-action audit logs. For many groups, buying a controlled runtime is safer than an under-governed internal prototype connected to the PMS.