Walt, your model gateway

Orchestrator · Layer 3

"Alright everyone, here's the plan."

What Walt does

Walt gives your tools one endpoint to call for AI. Behind it he decides which model answers, records what each request cost, and also builds and ships releases.

One endpoint

Apps and services send requests to Walt in a familiar messages-style API. They don't need to know which provider or local model sits behind it.

Provider-neutral routing

Requests go through the carousel to local models on Ollama or to cloud providers such as OpenRouter, Groq and Anthropic, picked by cost and capability.

Cost ledger

Every request is metered, so you can see what each call, project and provider actually cost.

A gated front door

Callers present a Grant-issued token or a shared ingress key. Walt uses his own upstream provider credentials, so callers never pass provider keys through.

Release builds

Walt builds a release from a specific commit, runs a smoke test and writes a manifest, so what you ship is what you tested.

Config-driven

Providers and routing rules live in config files. Change the plan by editing a file, not by rebuilding.

How it works

A request takes four steps, and Walt owns all of them.

  1. A caller sends a request with its credentials.
  2. Walt checks the credentials against Grant.
  3. The carousel picks a model by cost and capability, local or cloud.
  4. The answer goes back to the caller, and the cost goes into the ledger.

Where it fits in the four layers

1 Infrastructure & roles Vera, Grant, the bus, the device registry
2 Data pipeline & state Flo, DynaRAG, the carousel
3 Reasoning & inference Rick, Annie, Max, Hank
4 UI & collaboration Marco, Noted, Norm, the graph

Walt sits in layer 3. He doesn't call the other layers directly: Grant decides who may call him, and the cost and request events go out on the bus where Mona shows them.