Catalog API

The catalog API lists the models you can use and the voices of each text to speech model. Use it to build your own model or voice picker, or to check what an agent can use in a region before you create or move it. Send requests to any regional base URL, for example https://eu-west.api.slng.ai. Every region returns the same catalog, so the host does not filter the results. Use the query parameters instead. See Regions and base URLs for every host. Authentication is optional. Without a key you get the public catalog. Add Authorization: Bearer $SLNG_API_KEY to get the models and pricing available to your organization.

Find models for an agent

Add surface=agent to list the models enabled for agents. This example lists text to speech (TTS) models for an English-speaking agent in Germany (eu-west):

Request

curl --get "https://eu-west.api.slng.ai/v1/catalog/models" \
  -H "Authorization: Bearer $SLNG_API_KEY" \
  --data-urlencode "surface=agent" \
  --data-urlencode "agent_region=eu-west" \
  --data-urlencode "language=en" \
  --data-urlencode "service_type=tts" \
  --data-urlencode "page_size=1"

Response

{
  "items": [
    {
      "code": "deepgram/aura:2",
      "name": "Aura 2",
      "service_type": "tts",
      "languages": ["en"],
      "voice_count": 41,
      "voices": [],
      "agent_availability": [
        {
          "region": "eu-west",
          "language": "en",
          "voice_ids": ["aura-2-amalthea-en", "aura-2-andromeda-en"]
        }
      ]
    }
  ],
  "meta": { "page": 1, "page_size": 1, "total": 6, "pages": 6 }
}

Run the same request with service_type=stt and service_type=llm to find the other two models. Then set them on the agent:

Copy each model’s code into models.stt, models.llm, or models.tts.
For models.tts_voice, pick a voice ID from the agent_availability entry that matches the agent’s region and language. Speech to text (STT) and LLM entries have an empty voice_ids list.

LiveKit and Pipecat agents use the same models, so there is no orchestrator filter.

Query parameters

Parameter	Use it to
`surface=agent`	List only the models enabled for agents, each with `agent_availability`
`agent_region`	Pick an agent region, such as `eu-west` or `us-east-slng`. Requires `surface=agent`
`language`	Pick a language, such as `en`. Repeat it for several languages
`service_type`	Pick `stt`, `llm`, or `tts`. Omit it to list all three

agent_region takes the agent’s region code, not a gateway code. An unknown code returns an empty list. The API reference covers the other filters: provider, model hosting location, protocol, use case, capability, and search.

Get a model’s voices

The list response never includes voices. Get the model with the same filters to fill voices, for example to show voice names and previews in a picker:

Request

curl --get "https://eu-west.api.slng.ai/v1/catalog/models/deepgram/aura:2" \
  -H "Authorization: Bearer $SLNG_API_KEY" \
  --data-urlencode "surface=agent" \
  --data-urlencode "agent_region=eu-west" \
  --data-urlencode "language=en"

Response

{
  "code": "deepgram/aura:2",
  "name": "Aura 2",
  "service_type": "tts",
  "voice_count": 41,
  "voices": [
    {
      "voice_id": "aura-2-amalthea-en",
      "name": "Amalthea",
      "language": "en",
      "gender": "feminine",
      "preview_url": "https://assets.slng.ai/voice-previews/deepgram/aura:2/7e9d4fd4-b44d-401a-ae75-27824285519d-fcf9245a.mp3"
    }
  ]
}

With surface=agent, voices holds only the voices available for that region and language. The endpoint returns the model object directly, and 404 when the model is not visible to you or does not match the filters. See Get model for the full schema.

Page through results

Pass page and page_size, and request the next page until you reach meta.pages. The default page_size is 20 and the maximum is 100. With surface=agent, each model counts once toward page_size. In the general catalog, a page holds page_size model families, and each family brings its matching variants, so items can be longer than page_size. meta.total counts families there too.

Region fields

A model’s agent regions, hosting location, and serving regions can all differ. A model can be available to an agent in one region while its deployment runs in another.

Field	What it says
`agent_availability`	The agent regions and languages the model is available in, with voice IDs. Filled only with `surface=agent`
`available_regions`, `region_details`	Where the model’s deployments run. Filter on these codes with `region`
`served_regions`, `served_region_details`	The regional gateways that can serve the model, including ones that reach a deployment hosted elsewhere

Use agent_region to find models for an agent, and region to filter by hosting location. Prices and latencies in the region details are the best available from that region, not a quote for a particular request. A missing measurement is null.

Get started

Regions

Execution layer

Agent Builder

Models

Integrate

Find models for an agent

Query parameters

Get a model’s voices

Page through results

Region fields

Next steps

Agent regions

Create an agent

​Find models for an agent

​Query parameters

​Get a model’s voices

​Page through results

​Region fields

​Next steps

Agent regions

Create an agent

Find models for an agent

Query parameters

Get a model’s voices

Page through results

Region fields

Next steps