Configure the AIFast OpenAI-compatible API in Dify

Verify API before configuring

First confirm that the API Key and model list are available:

curl https://www.aifast.club/v1/models \
  -H "Authorization: Bearer $API_KEY"

Copy the real model ID from the response, do not directly fill in the manufacturer name, Chinese display name, or sample model in the article.

Add a model provider in Dify

Dify's menu name changes depending on the cloud service, community edition, and plug-in version, and is usually accessed from "Settings/Model Provider" in the workspace.

1. Open the model provider page.
2. Choose OpenAI Compatible, OpenAI API Compatible, or a vendor plug-in that supports custom endpoints.
3. Fill in the AIFast API Key.
4. Base URL fill in https://www.aifast.club/v1。
5. add from /v1/models The actual model ID returned.
6. Choose model types such as chat, embedding, and rerank based on actual capabilities.
7. Save and perform vendor-provided credential verification.

If the plug-in will automatically append /v1, should be filled in according to the final request log https://www.aifast.club, to avoid /v1/v1。

Minimal verification

Create a new workflow with only a start node and a large language model node:

1. Enter a canned question, such as "Reply OK only."
2. Temporarily close knowledge bases, tools, and complex prompts.
3. Verify that model nodes can return text.
4. Then gradually add knowledge bases, variables, structured output and tools.

The success of basic chat does not mean that Embedding, Rerank, tool calling and multi-modal capabilities are all available. These functions should be selected and tested separately.

Common mistakes

Credential verification failed

• Check whether the Key is complete.
• Confirm that the request header uses Bearer Token.
• Check if Base URL is repeated /v1/v1。

model not found

• Request /v1/models。
• Use the returned id, do not use the page display name.
• Checks whether the model is suitable for the current node type.

Workflow can talk but knowledge base fails

Knowledge bases often also require separate embedding models. Just because the chat model is available does not mean that the vector model has been configured.

When you need to configure vector dimensions, blocking, Rerank and retrieval acceptance, continue readingEmbedding, Rerank API and Dify knowledge base access tutorial。

429 or execution time is too long

Reduce concurrency, shorten context, check balances, throttling, and duplicate calls in workflows. Don't retry infinitely for 401, 400, and model name errors.