LLM Providers

An LLM Provider holds the connection credentials for a Large Language Model API. Providers are scoped per tenant, and you need at least one before any agent can run. The zango-app-developer plugin creates and configures providers for you through the platform API, so this page is a reference for what a provider is and what it stores, not a click-by-click guide.

You describe, the plugin provisions

Tell the plugin "use Anthropic Claude for this app" and it creates the provider, stores the key encrypted, and selects a default model through the platform API. The same settings are available to adjust by hand in the App Panel if you ever need to.

An LLM provider is configured once per tenant; its API key is stored encrypted and only the name is shown afterwards.

Supported providers

Zango currently supports OpenAI and Anthropic. Available models are fetched automatically from the provider once a valid API key is saved. Azure OpenAI and Amazon Bedrock are planned.

What a provider stores

Field	Description
Provider Type	OpenAI, Anthropic, or custom
API Key	Stored encrypted with field-level encryption; never exposed in plaintext after saving
Organization ID	OpenAI only; optional organization identifier
Model	The default model selected from the provider's available list
Rate Limit	Maximum requests per minute; blank for no limit
Monthly Budget (USD)	Spend cap; Zango stops routing requests once it is reached

Security

API keys are encrypted at rest using Zango's FIELD_ENCRYPTION_KEY. The key value is write-only: once saved, only its name is visible, and you can rotate it at any time.

tip

Never put LLM API keys in your app code or the project .env file. Provider credentials live encrypted per tenant and are not shared across apps. This is the same Secrets mechanism the rest of the platform uses.

Multiple providers

You can configure multiple providers per tenant, and each agent selects its provider individually. That lets you mix providers across agents in the same tenant, for example a fast, cheap model for classification and a frontier model for synthesis.

Next steps

With a provider in place, create an agent and select it.