Skip to main content

MiniMax

Reaches MiniMax's own API with a key from your MiniMax account.

This type is addressed using the Chat Completions protocol. That protocol has nowhere to put a server-side search tool, so web search is not available here: the Web column shows for every model.

The fields below are the standard provider fields. For what each one means on every provider type, see the provider field reference.

Before you start

  • A MiniMax account with API access.
  • An API key. Create one from the API Keys area of the MiniMax platform console. MiniMax also documents the OpenAI-compatible base URL your account should use, and the base differs by service region — check which applies to yours in MiniMax's quickstart. (Content was rephrased for compliance with licensing restrictions.)
  • The AI settings permission, and the exact MiniMax model ids you intend to use. Ids are typed here, not discovered.
Decide what may leave your environment

Prompts sent to this provider leave your installation and reach MiniMax. Confirm that is permitted under your organisation's agreements before you enable a model, and set the limits in Policy and Limits.

Set it up

  1. Go to Settings → AI → Setup and click Edit, then Add provider.
  2. Set Name to something you will recognise, for example minimax. It must be unique on the instance — the stored credential is keyed by it.
  3. Set Type to MiniMax. The card then states that it is addressed using the Chat Completions protocol, and that this protocol cannot carry web search.
  4. Click Route through a custom endpoint and set Endpoint to the full chat-completions URL for your MiniMax account, taking the base from the vendor's documented endpoint for your region and adding the chat-completions path.
  5. Leave Sign-in method on API key.
  6. Paste the key into Secret. It is stored encrypted and never shown again.
  7. Under Permitted models, type each MiniMax model id exactly as MiniMax names it, add a Display name if you want a friendlier label in the chat picker, and switch On for every model this instance may use. Set Cost and Auto as you want them. Web is not offered.
  8. Click Save.
  9. Click Test connection and read the row for each enabled model.
Enter the complete URL

For a chat_completions provider Linkiir sends the request to the Endpoint exactly as entered. Enter the full chat-completions URL for the service, not just its host. If Test connection reports Unreachable or Failed, the endpoint is the first thing to check.

MiniMax models are also available through Bedrock

If your AWS account is entitled to the MiniMax models on Bedrock, you can configure them there instead — see AWS Bedrock. That route uses your AWS credential rather than a MiniMax key, and is addressed by region.

Configuration reference

FieldWhat to set for this type
NameAnything unique on the instance. Keys the stored credential
TypeMiniMax
API shapeDerived as chat_completions. Change it only through Override, and only if your service speaks a different protocol
RegionNot shown. This type is not addressed by region — the region is part of the URL you enter
EndpointBehind Route through a custom endpoint. The complete chat-completions URL, sent exactly as entered
Sign-in methodAPI key (the default), or Client certificate where you terminate mTLS
SecretThe MiniMax API key. Stored encrypted; leave the mask in place to keep the existing one
Permitted models → Model idTyped verbatim, exactly as MiniMax names the model
Permitted models → WebShows . This protocol cannot carry web search

Context window, tool support and image support are not entered here — they ship with the release per model family. A model id the release does not recognise is treated conservatively: small context, no tools, no images. It is still offered for simple questions, but it is never picked for Build and never for an image.

Verify it worked

  • Test connection shows one row per enabled model, and every row reads OK.
  • The provider's collapsed row shows the model count and no "incomplete" badge.
  • In the chat model picker the models appear under the MiniMax vendor badge, with no internet indicator.

If it didn't work

Test connection saysCauseFix
UnreachableThe Endpoint is wrong, or the Linkiir server cannot reach itCheck the full URL first, then outbound network access and any proxy
FailedThe endpoint answered, but not with a chat-completions responseUsually a host or path that is not the chat-completions URL. Re-check the complete URL
IncompleteA required field is empty, or no model is enabledFix the issue named on the provider's row, then Save and re-test
No credentialThe Secret was never saved, or was clearedPaste the key again and Save before testing
Bad credentialThe key is wrong, or belongs to a different service regionRe-copy the key, and confirm the endpoint matches the region the key was issued for
Not allowedThe key is not permitted to call this modelCheck the key's permissions in the MiniMax console
Not enabled for accountThe account has no access to that modelUse a model the account is entitled to
Rate limitedThe account's rate or quota limit was reachedRetry later, or raise the limit with MiniMax
Unknown modelThe model id is misspelled or retiredRe-copy the id from MiniMax's model list. Ids are used verbatim
Save is refused, naming a model and web searchWeb was switched on before the type was set to MiniMaxTurn Web off — this protocol has no way to carry it
Results look staleTest connection tests the saved providerThe order is paste → SaveTest connection

Next