> ## Documentation Index
> Fetch the complete documentation index at: https://api-docs.ollang.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Provider Catalog

> How Ollang orchestrates AI providers, what each pipeline step does, and how to read provider status and order-type support.

Ollang is an **orchestration layer**, not a single-vendor pipeline. Every Order is assembled from independent provider choices — one per pipeline step — so you can mix a best-in-class transcription model with an efficient translation model and a premium voice, and change any one of them without rebuilding the workflow.

This catalog documents the providers that are **selectable in Workflows today** and what each is genuinely good at.

<CardGroup cols={2}>
  <Card title="Speech-to-Text" icon="waveform-lines" href="/providers/speech-to-text">
    8 selectable transcription providers, from multi-agent to single-model.
  </Card>

  <Card title="Translation and LLMs" icon="language" href="/providers/translation">
    27 translation engines and frontier LLMs across 10 model families.
  </Card>

  <Card title="Text-to-Speech" icon="microphone-lines" href="/providers/text-to-speech">
    6 voice synthesis engines for AI Dubbing and Audio Description.
  </Card>

  <Card title="Audio Processing" icon="scissors" href="/providers/audio-processing">
    Stem separation that preserves music and effects under dubbed dialogue.
  </Card>
</CardGroup>

***

## The four pipeline steps

Providers are grouped by the **step** they run in. An Order only runs the steps its Order Type needs.

```mermaid theme={null}
flowchart LR
  A["Source media"] --> B["split<br/>Audio Separation"]
  B --> C["transcribe<br/>Speech-to-Text"]
  C --> D["translate<br/>Translation"]
  D --> E["textToSpeech<br/>Voice Synthesis"]
  E --> F["Deliverable"]
```

| Step             | Internal key   | What it produces                                    | Used by                                |
| ---------------- | -------------- | --------------------------------------------------- | -------------------------------------- |
| Audio Separation | `split`        | Isolated dialogue stem plus a music-and-effects bed | AI Dubbing                             |
| Speech-to-Text   | `transcribe`   | Timecoded transcript with speakers                  | Closed Captions, Subtitles, AI Dubbing |
| Translation      | `translate`    | Localized text, subtitles, or dubbing script        | Documents, Subtitles, AI Dubbing       |
| Voice Synthesis  | `textToSpeech` | Synthesized target-language audio                   | AI Dubbing                             |

<Info>
  A **Document Translation** Order touches only `translate`. A **Closed Captions** Order touches only `transcribe`. An **AI Dubbing** Order can run all four, so its provider selection is really four decisions rather than one.
</Info>

***

## Reading a provider entry

Every provider in this catalog carries the same three attributes.

<AccordionGroup>
  <Accordion title="Provider tag" icon="tag">
    The stable machine identifier, for example `elevenlabs` or `anthropic/claude-opus-5`. This is the value returned by the [List Consumption](/apis/ollang-api-reference/list-consumption) endpoint and accepted by its `filter[provider]` parameter, so it is what you match on when you reconcile provider activity programmatically. Display names change; provider tags do not.

    One tag is reused across steps: `aws` identifies both Amazon Transcribe in the `transcribe` step and Amazon Translate in the `translate` step. Filtering on `aws` returns rows from both, so narrow with `filter[orderType]` or check the service field.
  </Accordion>

  <Accordion title="Order types" icon="list-check">
    Which Order Types can select the provider. Most translation engines serve `document`, `subtitle`, and `aiDubbing`, but there are real exceptions — DeepL is document and subtitle only, and Llama 4 is not available for dubbing scripts. See [Order Types](/apis/ollang-api-reference/order-types) for the full enum.
  </Accordion>

  <Accordion title="Status" icon="circle-info">
    <Tooltip tip="Selected automatically when no other provider is configured">**Default**</Tooltip> — the provider Ollang uses for that step when a Workflow does not override it.

    **New** — recently added and fully supported.

    **Experimental** — available for evaluation, behavior may change, and output should be reviewed before it goes to production.
  </Accordion>
</AccordionGroup>

***

## Defaults

If you never configure a Workflow, Orders run on these:

| Step             | Default provider                                                                   |
| ---------------- | ---------------------------------------------------------------------------------- |
| Audio Separation | [LALAL.AI](/providers/audio-processing)                                            |
| Speech-to-Text   | [Agentic STT](/providers/speech-to-text#agentic-stt)                               |
| Translation      | [Agentic Localization](/providers/translation#ollang-agentic)                      |
| Voice Synthesis  | [ElevenLabs Multilingual V2](/providers/text-to-speech#elevenlabs-multilingual-v2) |

These defaults favor quality and robustness. Each catalog page also identifies the efficiency-oriented alternatives that suit high-volume or lower-stakes content.

***

## Choosing a provider

<Steps>
  <Step title="Start from the content, not the benchmark">
    Leaderboard position rarely predicts localization quality for *your* material. Clean studio narration, noisy field audio, and code-switched conversation reward completely different models.
  </Step>

  <Step title="Check language coverage first">
    This is the most common reason a Workflow has to be reconfigured. Coverage across the catalog ranges from 7 languages to well over 100, and the highest-accuracy tier of a provider is often narrower than its headline language count.
  </Step>

  <Step title="Set a Global Workflow as your baseline">
    Configure one organization-wide default that is safe for most content, then override per Folder where a specific team or content type needs something different. See [Workflows and Provider Architecture](/workflows-provider-architecture).
  </Step>

  <Step title="Benchmark with parallel Folders">
    Run the same source through two Folders with different providers and compare AI QC scores and human edit distance. This is the only measurement that reflects your content and your reviewers.
  </Step>

  <Step title="Re-check quarterly">
    The catalog moves fast — several entries here did not exist six months ago. A provider decision that was correct last quarter is worth re-testing rather than assumed.
  </Step>
</Steps>

<Warning>
  Changing a Workflow only affects **future** Orders. Orders already created keep the providers they were created with, so a provider swap is never retroactive and never changes a deliverable you have already reviewed.
</Warning>

***

## Scope of this catalog

The pages in **STT APIs**, **Translation APIs**, **TTS APIs**, and **Audio Operations** also document engines from the wider localization ecosystem that Ollang can integrate with on request. Those reference pages are useful background, but only the providers listed in this catalog are selectable in Workflows without a custom integration.

If you need an engine that is documented but not in the catalog, contact your account team — adding a provider to a workspace is a configuration change, not a product change.