AI Providers

Settings > AI Providers is where the assistant gets its brain. You add one or more provider credentials, eChat uses the first working one, and you pay that provider directly for what your customers use.

You supply the API key

eChat does not resell model access and has no credit balance of its own. The key you add here is your own account with OpenAI, Anthropic, Google, or xAI, and the usage is billed to that account. No key is needed in the Magento admin - it belongs here. See Costs and billing.

Supported providers and models

eChat works with OpenAI, Anthropic, Google Gemini and xAI. The models you can pick from each are listed in the Model dropdown when you add or edit a credential - that list is the authoritative one, and it follows the version of eChat you are running.

The choice is curated rather than open. For each provider:

• The model marked (recommended) is the vetted floor: it is the default, and it is where eChat lands if a model you selected is no longer offered.
• Above it sit stronger models to opt up to, when answer quality matters more than cost.
• A few faster, cheaper models are offered as a deliberate trade of reasoning depth for speed. They are never the default - pick one only if you have tried it against your own catalogue.
• Below that, micro tiers, older generations and specialized variants are not offered at all: the assistant stops being reliable enough to put in front of customers.

Model identifiers rotate quickly as providers release new versions, which is why they are not reprinted here. A model dropped from the list is not a broken store - eChat falls back to that provider's recommended model rather than failing.

Adding a provider

Click Add provider and fill in:

Field	Description
Alias	Your own name for this credential, for example openai-main or claude-fallback. It identifies the credential in the list and in status messages.
Provider	OpenAI, Anthropic, Google Gemini, or xAI.
API Key	The key from your provider account. It is stored encrypted and never shown again - the list displays a masked form. When editing an existing credential, leave the field blank to keep the stored key, and enter a new key only to replace it.
Model	The model this credential runs on. The dropdown lists what this eChat version offers for the chosen provider; it defaults to the recommended one.
Enabled	Whether this credential takes part in answering.

A new credential requires a key; an existing one does not, so you can change its model without re-entering the key.

The fallback chain

Credentials are ordered, and the order is the fallback chain. eChat uses the first enabled, working credential. If it is unavailable - out of quota, rate limited, key revoked - eChat moves to the next one automatically, and the conversation continues.

Use the move controls on each row to reorder the chain. A sensible setup is one provider you intend to run on, plus one from a different vendor underneath it, so an outage at a single vendor does not take your chat down.

If every credential fails, the assistant is disabled. eChat emails you about it only if Email me when the AI chat stops working is on, and that setting is off by default - switch it on once, on the AI Assistant tab. Without it, a chain that has run out of working credentials stops answering silently.

Status and revalidation

Each credential carries a status, set when eChat last probed it or last used it - a conversation that fails updates the status of the credential it failed on, which is also what moves traffic down the chain. Revalidate re-checks it on demand.

Status	Meaning	What to do
OK	The key and model answered normally.	Nothing.
UNCHECKED	The credential has not been probed yet.	Click Revalidate.
EMPTY_KEY	No key is stored for this credential.	Edit the credential and enter the key.
INVALID_KEY	The provider rejected the key.	The key was revoked, mistyped, or belongs to a different project. Issue a new one and replace it.
INSUFFICIENT_QUOTA	The key is valid but the account has no credit.	Top up the provider account. This is the most common cause of a chat that worked yesterday and does not today.
RATE_LIMITED	The provider is throttling this key.	Usually the account's usage tier is too low for your traffic. Raise the tier with the provider, or put a second provider under this one in the chain.
MODEL_UNAVAILABLE	The selected model is not available to this key.	Some models require a specific plan or verification with the provider. Select the recommended model, or unlock the model on the provider side.
TRANSIENT_ERROR	A temporary failure at the provider.	Revalidate. If it persists, the fallback chain has already routed traffic onwards.

See also Troubleshooting.