Agents Catalog

Source list for the agents shown on registry.inference-gateway.com.

This repo doesn't store agent definitions itself — each agent's ADL agent.yaml lives in its own GitHub repo and is the source of truth. This repo just lists which agents to include in the catalog. A scheduled build job pulls each agent.yaml, validates it against the ADL JSON Schema, and bundles everything into a single catalog.json served via jsdelivr.

Layout

agents.yaml                 # the only file you edit — list of GitHub repos to include
scripts/build-catalog.mjs   # fetches + validates + bundles into catalog.json
catalog.json                # generated, committed; consumed by the registry SPA
.github/workflows/build-catalog.yml

Adding an agent

1. Make sure the agent's repo has an ADL agent.yaml at its root (see inference-gateway/adl for the schema).

2. Open a PR appending one entry to agents.yaml:

   - url: https://github.com/some-org/cool-agent
     ref: main # branch, tag, or SHA. Pin to a release tag if you can.

3. On merge, CI rebuilds catalog.json. The live registry picks it up within the jsdelivr @main cache window (up to ~12h).

Third-party agents are welcome — the url does not have to live under inference-gateway.

Catalog endpoint

https://cdn.jsdelivr.net/gh/inference-gateway/agents@main/catalog.json

Shape:

{
  "version": 1,
  "updated": "2026-05-25T00:00:00Z",
  "agents": [
    {
      "apiVersion": "adl.inference-gateway.com/v1",
      "kind": "Agent",
      "metadata": { "name": "...", "description": "...", "version": "..." },
      "spec": { "...": "..." },
      "_source": {
        "url": "https://github.com/...",
        "ref": "main",
        "fetchedAt": "..."
      }
    }
  ]
}

Each agent doc is a pure ADL manifest plus a non-schema _source block recording where the aggregator pulled it from.

Local build

npm install
npm run build      # writes catalog.json

The build script:

1. Reads agents.yaml.
2. Fetches agent.yaml from raw.githubusercontent.com/<owner>/<repo>/<ref>/agent.yaml for each entry.
3. Validates against the ADL JSON Schema (adl/schema/v1/schema.json) via Ajv.
4. Rejects duplicate metadata.name collisions.
5. Sorts by metadata.name and writes catalog.json.

Override the ADL schema location with ADL_SCHEMA_URL=... if needed (for testing against a fork).

Schedule

The workflow runs on push to agents.yaml / scripts / workflow, plus a daily cron (0 4 * * * UTC) so upstream agent.yaml version bumps roll into the catalog without manual intervention.