23 — embedded URIRUN layer

Every other example installs a connector to add routes. This one shows the routes urirun carries *by itself* — the embedded layer that resolves with no connector installed and no registry file. The runtime answers them directly:

Both are *resolver builtins* (_builtin_registry_route_entry / _builtin_error_route_entry in runtime/v2.py): when a registry:// or error:// URI is not in the registry, the runtime mounts it automatically. So the introspection and observability surface uses the same URI contract an LLM/MCP/A2A client already speaks.

> log:// (runtime logs) is often grouped with these, but it lives in the > host layer and needs host setup, so it is intentionally left out of this > zero-config demo.

Run

./demo.sh

Requires urirun ≥ 0.4.4 (the registry:// builtin and zero-config run). Override the binary if needed:

URIRUN="python3 -m urirun.v2" ./demo.sh

What it does, all without a connector or a compiled registry:

1. registry://local/routes/query/list — list every live route in the runtime.
2. same route, filtered locally — show just the embedded registry:///error:// layer.
3. registry://local/bindings/query/show — inspect one binding's full contract.
4. error://local/errors/query — query the runtime error store.

Why this matters

The runtime is self-describing and self-observing over URIs. You do not need a special API to ask "what can I run?" (registry://) or "what went wrong?" (error://) — they are routes like any other, discoverable in urirun list and runnable with urirun run. Any failed run is recorded with a stable error://local/<code>/query/info address you can look up later.

See also: 13-simple_defaults for the connector *authoring* side (declaring your own routes).

Files

View on GitHub →