Spettro Provider Catalog

Curated catalog of LLM providers and chat models that are verified to work with Spettro. Built automatically and published to catalog.spettro.app.

Published files (the only two files served):

• https://catalog.spettro.app/providers.json — pretty-printed, for humans
• https://catalog.spettro.app/providers.min.json — minified, what the Spettro CLI downloads

The catalog is rebuilt on every push to main and every six hours (via GitHub Actions cron) to pick up new models from models.dev.

Format

{
  "version": 1,
  "updated": "2026-07-13",
  "providers": {
    "groq": {
      "name": "Groq",
      "api": "openai",
      "base_url": "https://api.groq.com/openai/v1",
      "env": "GROQ_API_KEY",
      "models": {
        "llama-3.3-70b-versatile": {
          "name": "Llama 3.3 70B",
          "tool_call": true,
          "context": 131072
        }
      }
    }
  }
}

• api is "openai" or "anthropic" — the wire protocol Spettro uses to talk to the provider. These are the only two protocols Spettro supports.
• base_url is the API endpoint. For api: "openai" it is the OpenAI-compatible root, usually ending in /v1. For api: "anthropic" it is the root without /v1 (the Anthropic client appends /v1/messages itself).
• env is the environment variable Spettro reads the API key from.
• Model flags (reasoning, tool_call, vision) are omitted when false; context is the context window in tokens; status (alpha/beta) is omitted when stable. Only chat models are included — embeddings, image/audio generation, rerankers etc. are filtered out.

Contributing

Add models/providers sourced from models.dev

Curated providers come from models.dev, filtered by config.json. To propose a new one, open a PR adding its models.dev id to the curated map with at least a base_url override (and api/env if needed). Please state in the PR that you actually tested the provider with Spettro.

Add a custom provider (your own endpoint, not on models.dev)

Add a file custom/<your-provider-id>.json following custom/_example.json: id, display name, api (openai or anthropic), base_url, env, and the list of models with their capabilities. Files starting with _ are ignored by the build.

CI validates every PR by running the full build; a provider with a missing field, an invalid api value, or a duplicate id fails the check.

Building locally

node scripts/build.mjs          # writes dist/providers.json + dist/providers.min.json
# offline (reuse a downloaded models.dev dump):
MODELS_DEV_JSON=/path/to/api.json node scripts/build.mjs

Requires Node 18+, no dependencies.

Hosting

Deployed to GitHub Pages via .github/workflows/deploy.yml. One-time setup:

1. Repo Settings → Pages → Source: GitHub Actions.
2. Set the custom domain to catalog.spettro.app and add a DNS CNAME record pointing catalog.spettro.app to <user>.github.io.