Supported Providers
Agent Router routes to 200+ models across the major commercial providers, the hyperscaler AI platforms, and self-hosted or open-model backends through a single OpenAI-compatible endpoint. This page lists the provider families that can be connected, how each one is authenticated, and where the authoritative, always-current list of routable models is published. Providers are configured once in the Admin Dashboard by a platform operator; developers then route to any enabled model without handling provider credentials directly.
How providers are connected
A provider is an upstream AI service the gateway is permitted to reach. Each provider carries its own credentials and connectivity, and each exposes a set of models that a platform operator enables for developers. The connection style differs by provider:
| Connection style | What is supplied | Example providers |
|---|---|---|
| API key | A single secret token | OpenAI, Anthropic, Mistral, Cohere, Together, Perplexity, Groq, xAI |
| Cloud credentials | A regional endpoint plus cloud IAM or a deployment key | Azure OpenAI, AWS Bedrock, Google Vertex AI |
| Native API key | A provider-issued key for the provider's own API surface | Google Gemini |
| Custom endpoint | A base URL for a self-hosted or compatible service | Self-hosted and OpenAI-compatible backends |
Provisioning, credential rotation, and model enablement are covered in Provision models and providers. AWS Bedrock has region-specific requirements documented separately in Provision AWS Bedrock models.
Provider families
OpenAI
GPT model family. Connected with an API key.
Anthropic
Claude Opus, Sonnet, and Haiku families. Connected with an API key.
Google Gemini
Gemini models through the native Google AI API.
Google Vertex AI
Gemini and partner models through a Google Cloud project.
Azure OpenAI
Azure-hosted OpenAI deployments with enterprise Azure credentials.
AWS Bedrock
Bedrock-hosted models with region-specific endpoints and IAM.
Mistral
Mistral model family. Connected with an API key.
Cohere
Command models and Cohere embeddings. Connected with an API key.
Together
Open-source models hosted on Together AI. Connected with an API key.
Perplexity
Search-augmented models. Connected with an API key.
Groq
Low-latency inference for Llama, Qwen, and open models.
Self-hosted
OpenAI-compatible and custom backends reached at a configured base URL.
The provider set grows over time, and the dropdown in Providers Management in the Admin Dashboard is the authoritative list of provider types available in a given release.
The authoritative model list
Individual model identifiers, context windows, and pricing tiers change frequently as providers ship new versions. Rather than duplicate that list here, two live sources hold the current state:
- Console Model Catalog: the models a platform operator has enabled for a deployment, visible to developers in the Console.
- Public models endpoint: the machine-readable catalog of routable models, published at
https://router.tetrate.ai/api/public/models.
Only models that a platform operator has explicitly enabled are routable. A model offered by a connected provider is not reachable until it has been enabled in Models Management.
Provider translation
Applications send requests in one of the supported API formats, and the gateway translates transparently to each provider's native API. An OpenAI Chat Completions request that routes to Anthropic Claude is converted to the Anthropic Messages format before it is forwarded, and the response is converted back. No per-provider client code is required; the base URL and API key are the only application-side changes.
For provider-side concerns that fall outside the gateway (model pricing, rate-limit tiers, and account management), each provider's own documentation is the canonical source. See External links for the current documentation URLs.
Related
- Provision models and providers: the operator workflow that connects a provider and enables its models
- Supported APIs: the request formats that route to every provider
- Gateway behavior: how requests are resolved, normalised, and routed across providers
Where to go next