WILDROADInternational

Business line

AI Agents & Automation

We are not an AI vendor talking about AI. We run Claude Code (Anthropic) as the daily operating layer of our own travel group, and we have built a private library of dozens of internal skills against it. This page describes what is actually running today.

Claude Code
Anthropic — our daily operating layer
Dozens of skills
Private wr-* skill library
Lark · CLI · API
Multi-surface, self-hosted

The stack we actually run

  • Claude Code · the daily driver

    Anthropic's Claude Code is our primary AI surface. Founder and ops run it as a terminal copilot against real production code, hotel data, order tables, and the company's internal docs.

  • Hermes · the private agent

    A long-running founder-side agent built on Claude Code. Handles email triage, Lark messaging, cron-driven reports, and personal workflow plumbing — the layer where new skills are first prototyped.

  • Xiaoye (小野) · the company agent

    Lark-facing bot for WR colleagues, running on OpenClaw with a curated read-only subset of our skills. Same brain, controlled blast radius.

  • wr-* skill library

    Internal skill repos covering hotel price queries, order auditing, supplier promotion sync, EDM campaigns, SEO auditing and more. Each skill is versioned in git and installable into either Hermes or Xiaoye.

Live use cases inside WR

  • Hotel price & rate-plan auditing

    Continuous price scans against suppliers; mismatched or stale rates flagged before they reach the booking flow.

  • Order ops & discount approvals

    A retainer-style audit agent reviews pre-paid GDS discount requests daily, escalating only the exceptions to a human reviewer.

  • Supplier promotion sync

    Partner promotion calendars are ingested into WR-side tags and rate plans automatically — chain-aware mapping for IHG, Marriott and others.

  • EDM, SEO & content pipelines

    Drafts, translation, structured publishing for itineraries and EDM campaigns; SEO audits across our three travel properties run as scheduled agent tasks.

Why this matters for partners

Because we built this for ourselves first, what we offer is not a slide deck or a chatbot demo. It is the same agent stack, the same skills, and the same operational discipline — installable into your team's workflows with observability and rollback from day one.

Open source

Most of our internal skill repos stay private because they contain production endpoints and credentials. What we can open-source, we do: