Dify with API429: OpenAI-API-compatible plugin fields, local limits and paid Save/Test

Verify your Dify OpenAI-API-compatible plugin version and scope

In your Dify workspace Settings under Model Providers, locate the official OpenAI-API-compatible plugin and add a custom LLM entry. The upstream plugin source reviewed on the verification date is version 0.0.69 on the main branch of langgenius/dify-official-plugins; your installed Dify deployment may run a different plugin revision. Compare the visible form controls in your workspace against the schema fields below before entering a live credential.

Scope the initial setup to a single plain-text chat completion model without knowledge base retrieval (RAG), agent tool calling or workflow loops. Although the OpenAI-API-compatible plugin form lists several model categories and advanced toggles, the presence of a checkbox in Dify does not establish backend support on API429. Embeddings, reranking, vision inputs, Responses API and agent workflows each require separate contract verification.

Enter the endpoint URL, exact model ID and strict Chat Completions fields

Enter your API429 credential into the secret-input api_key field, set endpoint_url to https://gateway.api429.com/v1 and set endpoint_model_name to the exact identifier gemini-3.8-flash. Do not confuse the wire model identifier in endpoint_model_name with an optional display alias. Confirm that gemini-3.8-flash is listed in the API429 public catalog and verify your specific key entitlement via an authenticated GET /v1/models request, keeping the secret out of screenshots and logs.

Select mode: chat and api_type: chat_completions so the plugin targets https://gateway.api429.com/v1/chat/completions rather than the Responses endpoint. Set token_param_name explicitly to max_tokens instead of leaving parameter selection on automatic detection, and set compatibility_mode to strict so the plugin omits unverified extended parameters. Do not enable reasoning toggles, function calling schemas or structured output flags during this initial connection check.

Set conservative local form limits without confusing them with native specs

In the field reference sheet below, context_size is set to '4096' and max_tokens_to_sample is set to '256'. These two numbers are conservative local Dify plugin settings chosen for a bounded first text check; they are not the native context window or maximum output specifications of Gemini 3.8 Flash. Inside Dify, context_size informs the application how much prompt context to permit locally, while max_tokens_to_sample sets the upper slider bound for completion tokens. Also set max_tokens to 256 inside your test app model parameters.

The JSON block below is a reference map of the Dify plugin form fields rather than an importable Dify DSL file, and YOUR_API429_KEY remains a literal placeholder. Starting with small local settings prevents oversized initial requests while you verify connectivity. If you later adjust context_size for longer prompts, base that decision on verified API429 limits for your key rather than assuming a Dify form input increases backend capacity.

Treat Save/Test credential validation as a paid live request

Clicking Save or Test in the Dify model modal is not a free offline syntax check. In the reviewed llm.py implementation, validate_credentials sends a live completion request to the configured endpoint and includes fallback retry attempts when certain token or parameter errors occur. Because a single Save/Test click can trigger one or more billable calls (or consume rate limits on failure), double-check endpoint_url, endpoint_model_name and api_key before submitting the form, and never click Save/Test repeatedly in a tight loop.

Once the credential saves cleanly, create a minimal blank chat app without tools or knowledge bases, select gemini-3.8-flash, disable streaming if your app view offers a non-streaming check and send one short nonconfidential question. Compare the returned answer and token usage in your API429 account dashboard. Keep credential validation and your first manual chat prompt distinct in your cost accounting, as both interact with the live model endpoint.

Control workspace access and monitor total deployment spend

A model credential saved under Dify Model Providers is shared across workspace members who have permission to build or publish apps using that provider. Configure Dify workspace roles carefully so unauthorized users cannot attach the shared API429 key to high-volume workflows or public web apps. Setting max_tokens_to_sample or max_tokens to 256 caps only a single completion output, not your total budget: input context, Save/Test validation calls, fallback attempts and multi-step app runs all contribute to total spend. Check current token rates on the model card and pricing comparison pages.

OpenAI-API-compatible plugin form field sheet; not a Dify import file

{
  "api_key": "YOUR_API429_KEY",
  "endpoint_url": "https://gateway.api429.com/v1",
  "endpoint_model_name": "gemini-3.8-flash",
  "mode": "chat",
  "api_type": "chat_completions",
  "token_param_name": "max_tokens",
  "compatibility_mode": "strict",
  "context_size": "4096",
  "max_tokens_to_sample": "256"
}

Target Chat Completions request contract; not a Dify app DSL file

{
  "model": "gemini-3.8-flash",
  "messages": [
    {
      "role": "user",
      "content": "Answer in one sentence: what is an API?"
    }
  ],
  "max_tokens": 256,
  "stream": false
}

Payment and a $10 test

• Russian bank card. Top up your balance in the account.
• Invoice for companies. Bank transfer with closing documents.
• USDT and TON. Top up in crypto.

$10 test credit. Subscribe to our Telegram channel through the bot and get $10 on your balance for the first requests. Get $10 in the bot