Simple pricing.
No token markups.

Never. When you route a request to Anthropic, OpenAI, or Gemini, the call goes directly to their API using your API key. You pay them. We charge zero on top of that. Dream-Weaver's pricing covers the gateway infrastructure itself, not the tokens.

Dream-Weaver runs in your infrastructure — your Kubernetes cluster, your Docker host, your server. We never touch your traffic. There's no Dream-Weaver cloud middleman. Your data stays in your network. The Pro license key just enables feature flags in your local install.

Yes. Cancel anytime via Stripe — no questions, no lock-in period. You keep Pro features until the end of your billing period. After that, you revert to Open Core features, which are still fully functional for local inference and explicit routing.

No. Without a GPU, you can still route to cloud providers (Anthropic, OpenAI, Gemini, etc.) and get everything except local inference. A GPU (NVIDIA or AMD) is what gets you the 60–80% API cost reduction by running the cheap/fast tasks locally. Most modern RTX cards (12GB+ VRAM) work great.

Kubernetes 1.26+ is recommended. The Helm chart works with k3s, k3d, GKE, EKS, AKS, and vanilla kube. You can also run via Docker Compose without any Kubernetes at all. See the download page for system requirements.

After subscribing, you receive a license key (format: DW-PRO-XXXX-XXXX-XXXX) by email. Pass it to the Helm chart via --set license.key=DW-PRO-... or set DW_LICENSE_KEY in your Docker Compose environment. The proxy validates the key on startup and enables Pro feature flags. No data leaves your cluster for license validation.

A lightweight mixture-of-experts classifier that reads each incoming request and dispatches it to the optimal backend in <5ms. It categorizes requests by intent (code, reasoning, long-context, creative, fast-lookup) and applies affinity scores across your configured backends. Local GPU first for cheap tasks, frontier models for the hard ones. You configure the policy; the classifier makes the call.