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

# Providers

> Every LLM provider TrustGate can route to, and which modalities each adapter supports — chat, embeddings, rerank, files, images, and audio.

<p className="nt-lead">
  A registry picks one of these adapters. Chat is always available. Embeddings,
  rerank, files, images, and audio only run when the adapter advertises them
  and the model slug can do that work.
</p>

Connect a card from **Agent Gateway → Registry**. How credentials and health
checks work is on [Registries](/trustgate/concepts/registries). Clients still
call one TrustGate path — [chat](/trustgate/endpoints/chat),
[embeddings](/trustgate/endpoints/embeddings),
[images](/trustgate/endpoints/images), [files](/trustgate/endpoints/files),
[audio](/trustgate/endpoints/audio), or [rerank](/trustgate/endpoints/rerank) —
and TrustGate adapts the wire format.

MCP servers are a different registry type. See the
[MCP Gateway](/trustgate/mcp/overview).

## Capability matrix

`Code` is the registry provider id (`owned_by` on
[`GET /v1/models`](/trustgate/endpoints/models), and the `@provider/…` prefix in
[model resolution](/trustgate/routing/model-resolution)).

### Text

| Provider                   | Code                | Chat | Embeddings | Rerank |
| -------------------------- | ------------------- | ---- | ---------- | ------ |
| OpenAI                     | `openai`            | Yes  | Yes        | —      |
| Anthropic                  | `anthropic`         | Yes  | —          | —      |
| Google Gemini              | `google`            | Yes  | —          | —      |
| Azure OpenAI               | `azure`             | Yes  | Yes        | —      |
| Amazon Bedrock             | `bedrock`           | Yes  | Yes        | —      |
| Vertex AI                  | `vertex`            | Yes  | Yes        | —      |
| Cerebras                   | `cerebras`          | Yes  | —          | —      |
| Cohere                     | `cohere`            | Yes  | Yes        | Yes    |
| DeepSeek                   | `deepseek`          | Yes  | —          | —      |
| Groq                       | `groq`              | Yes  | —          | —      |
| Mistral                    | `mistral`           | Yes  | Yes        | —      |
| OpenRouter                 | `openrouter`        | Yes  | —          | —      |
| xAI                        | `xai`               | Yes  | —          | —      |
| Custom / OpenAI-compatible | `openai_compatible` | Yes  | Yes        | —      |

### Files, images, and audio

| Provider                   | Files | Images | Speech | Transcription |
| -------------------------- | ----- | ------ | ------ | ------------- |
| OpenAI                     | Yes   | Yes    | Yes    | Yes           |
| Anthropic                  | Yes   | —      | —      | —             |
| Google Gemini              | —     | —      | —      | —             |
| Azure OpenAI               | Yes   | Yes    | Yes    | Yes           |
| Amazon Bedrock             | —     | —      | —      | —             |
| Vertex AI                  | —     | —      | —      | —             |
| Cerebras                   | —     | —      | —      | —             |
| Cohere                     | —     | —      | —      | —             |
| DeepSeek                   | —     | —      | —      | —             |
| Groq                       | —     | —      | Yes    | Yes           |
| Mistral                    | Yes   | —      | Yes    | Yes           |
| OpenRouter                 | Yes   | Yes    | Yes    | Yes           |
| xAI                        | Yes   | —      | —      | —             |
| Custom / OpenAI-compatible | —     | Yes    | Yes    | Yes           |

Yes means the **adapter** can forward that modality. A request still needs a
model that actually supports it — for example Bedrock embeddings only accept
Titan embed slugs, and OpenAI image calls need an image model.

Pinning a registry that cannot serve the path is a **400**, not a failover. An
empty capable pool is a **503**. See each endpoint page for the candidate pool.

## What each capability maps to

| Capability        | Client path                                                                                                                       | Notes                                                                                                     |
| ----------------- | --------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------- |
| **Chat**          | [`/v1/chat/completions`](/trustgate/endpoints/chat), `/v1/messages`, `/v1/responses`, Cohere `/v2/chat`, Gemini `generateContent` | Every provider. Inbound format is independent of the upstream.                                            |
| **Embeddings**    | [`/v1/embeddings`](/trustgate/endpoints/embeddings)                                                                               | OpenAI-shaped body. Gemini AI Studio (`google`) does not embed — use Vertex. Bedrock is Titan embed only. |
| **Rerank**        | [`/v1/rerank`](/trustgate/endpoints/rerank)                                                                                       | Cohere only.                                                                                              |
| **Files**         | [`/v1/files`](/trustgate/endpoints/files)                                                                                         | Upload, list, retrieve, delete, download. Custom OpenAI-compatible backends do not expose a files store.  |
| **Images**        | [`/v1/images/generations`](/trustgate/endpoints/images)                                                                           | Plus multipart edits and variations where the upstream accepts them.                                      |
| **Speech**        | [`/v1/audio/speech`](/trustgate/endpoints/audio)                                                                                  | TTS. Raw audio bytes.                                                                                     |
| **Transcription** | [`/v1/audio/transcriptions`](/trustgate/endpoints/audio)                                                                          | STT. Multipart.                                                                                           |

## Credentials

The credential on the registry is how TrustGate calls the **provider**. It is
not the [consumer key](/trustgate/concepts/auth) your applications send.

| Provider                                                                              | Auth                                                                        |
| ------------------------------------------------------------------------------------- | --------------------------------------------------------------------------- |
| OpenAI, Anthropic, Gemini, Mistral, Groq, Cohere, DeepSeek, xAI, Cerebras, OpenRouter | API key                                                                     |
| Custom / OpenAI-compatible                                                            | API key and/or a custom header pair, plus a `base_url`                      |
| Azure OpenAI                                                                          | API key, service principal, or managed identity, plus the resource endpoint |
| Amazon Bedrock                                                                        | Access keys (optional session token) or assume-role / IRSA, plus region     |
| Vertex AI                                                                             | GCP service-account JSON                                                    |

Use **Custom / OpenAI-compatible** when the upstream speaks Chat Completions
(and optionally `/v1/embeddings`, images, or audio) but is not a first-class
card — vLLM, Ollama, LiteLLM, and similar.

## Related

* [Registries](/trustgate/concepts/registries) — connect, test, and price a backend
* [Chat](/trustgate/endpoints/chat)
* [Model resolution](/trustgate/routing/model-resolution)
* [Models](/trustgate/endpoints/models)