SHOWCALL / DOCUMENTATION

Agent Catalog

Catalog every AI agent in your Backstage instance — across kagent, ARK, and any A2A service — with live agent cards and LLM-gateway usage.

Start here Full repository README

01 / QUICKSTART

Get the catalog running

The fastest path is the disposable local demo. It installs kagent by default, creates representative workloads, and starts a Backstage app wired to the local plugins.

TERMINALminikube / kind / k3d

minikube start --cpus=4 --memory=8192
./demo/check.sh
./demo/up.sh
./demo/backstage.sh

Open http://localhost:3001/agents after Backstage starts. Add ARK to prove multi-runtime discovery with DEMO_RUNTIMES="kagent ark" ./demo/up.sh.

Read the full demo guide Frontend package guide Backend package guide

02 / ENTITY MODEL

Agents become catalog citizens.

Agent Catalog uses standard Backstage kinds so existing ownership, search, scorecard, and orphan workflows keep working.

COMPONENT

Agent runtime

Owner, lifecycle, runtime, cluster, reachability, and flat annotations.

API

Live A2A card

What the running agent actually serves, fetched from the live endpoint.

RESOURCE

Model config

Model provider and configuration dependencies connected to the agent.

Read the concepts primer

03 / LOCAL DEMO

See the aha moment locally.

Known agents, shadow workloads, and unattributed LLM usage land in one Backstage-native view. No real model key is required.

Agent Catalog sample fleet view across multiple runtimes and discovery paths

Demo README Architecture

04 / ARCHITECTURE

Observe. Enrich. Govern.

01Runtime sources

CRDs, labeled Services, workload signals, and gateway usage.

02Catalog providers

Transform facts into Components, APIs, Resources, and relations.

03Backstage workflows

Ownership, search, fleet visibility, scorecards, and governance.

The catalog is deliberately outside the deploy path. It observes what runs; it does not run agents.

Read the architecture decision

05 / ENTERPRISE SUPPORT

Bring the hard environment.

Showcall can help teams map multiple runtimes and clusters, integrate gateway observability, review RBAC, and establish the operating model around agent discovery.

Talk to Showcall