The pitch on zymr.com is clear: "AI-first mindset shapes how we build. Intelligence is embedded at the core, not added later." ZOEY is the orchestration framework. Every client delivery now layers Generative AI, AI Agents, MLOps, and Data Engineering on top of it.
Each one of those deliveries calls a model. Often several. Sometimes from multiple providers. Always with client data flowing through. That's the layer where Cloudflare slots in — underneath ZOEY, underneath every Zymr-built AI workflow.
The pieces that map cleanly to what Zymr is already building for clients:
AI Gateway — under every ZOEY agent call. Caching, fallback routing, PII redaction, spend caps before the model provider sees the request.
Workers AI — for the inference path where Zymr clients don't want their data sitting in a third-party model provider's logs.
Vectorize + R2 — for the RAG layer that every "AI-native platform engineering at scale" engagement needs.
Durable Objects — for stateful agent orchestration, per-session, per-tenant.
One more piece worth knowing: zymr.com is already fronted by Cloudflare today (Webflow + Cloudflare delivery). The foundation is in place; the developer-platform conversation is the natural next layer.
Is the bigger pain point right now on the ZOEY-side — making your own agent framework production-ready for client deployments — or on the client-delivery side — needing infrastructure primitives that don't lock the client to AWS/Azure/GCP? 20 minutes to compare notes.
Sent as a follow-up to our LinkedIn connect. The detailed primitive-by-primitive mapping (ZOEY architecture × Cloudflare developer platform, plus the AI Gateway runtime story) is the natural second conversation if this thread has legs.