LLM providers

An LLM provider is a configured upstream model provider — OpenAI, Anthropic, or any service that speaks one of the protocols the platform supports. Each provider has an endpoint, an auth method, and the credentials needed to call it. You register a provider once per organization; then you register one or more models that point to it.

Supported protocols

The LLM Proxy speaks both protocols natively. Agents use whichever protocol corresponds to the model they call — no per-agent configuration is required.

Auth methods

For custom_headers, you cannot set Host, Content-Length, Connection, or Transfer-Encoding — these are managed by the proxy.

Register a provider

In the Console

1. Console → LLM Providers (/organizations/<org>/llm-providers).
2. Click New provider.
3. Fill in:
   • Name — display name (e.g. openai-prod).
   • Endpoint — base URL of the upstream API (e.g. https://api.openai.com/v1).
   • Protocol — responses or anthropic_messages.
   • Auth method — bearer, x_api_key, or custom_headers.
   • Credentials — the API key (for bearer / x_api_key) or a header map (for custom_headers). Credentials are masked after save; click the eye icon to reveal.
4. Save.

The Console hides the credential value after the first save. To rotate it, click Edit, paste the new key, and save again.

With Terraform

resource "agyn_secret" "openai_key" {
  organization_id = agyn_organization.acme.id
  name            = "openai-api-key"
  value           = var.openai_api_key
}

resource "agyn_llm_provider" "openai" {
  organization_id = agyn_organization.acme.id

  name        = "openai-prod"
  endpoint    = "https://api.openai.com/v1"
  protocol    = "responses"
  auth_method = "bearer"

  token_secret_id = agyn_secret.openai_key.id
}

For an Anthropic provider with the same model:

resource "agyn_llm_provider" "anthropic" {
  organization_id = agyn_organization.acme.id

  name        = "anthropic-prod"
  endpoint    = "https://api.anthropic.com/v1"
  protocol    = "anthropic_messages"
  auth_method = "x_api_key"

  token_secret_id = agyn_secret.anthropic_key.id
}

For custom_headers, pass a map under custom_headers instead of token_secret_id.

Test a provider

The provider page in the Console shows the count of models that use the provider. To verify the credentials work, use the Test action on any model registered under the provider — see Models.

Edit and delete

Editing a provider's credentials applies to future LLM calls. Calls already in flight finish with the old credentials.

Deleting a provider removes the configuration. Any models pointing at it are orphaned and stop resolving — agents using those models start failing their next LLM call. The Console warns you about dependent models before allowing the delete.

Self-hosted endpoints

For self-hosted or air-gapped deployments, you can register a provider pointing at any URL that speaks the responses or anthropic_messages protocol — for example, a vLLM or TGI server with the appropriate adapter in front. Set endpoint to your service URL and auth_method to whatever your gateway expects.

Related

• Models — give each provider one or more named models.
• Secrets — where the credential value lives.
• Operate → Architecture overview — how LLM Proxy bridges agents to providers.