Dify logo
Bot builders

Use TokenRex with Dify

Build Dify agents and chatbots with TokenRex through the OpenAI-API-compatible model plugin.

Choose a model with tool calling when building an Agent app. Set any required context and output limits from the selected model's verified details. Chat-model setup does not configure embeddings or document retrieval. Provider setup checked against the official reference. Live execution inside this third-party tool has not been tested; model limits, streaming and tool behavior must be verified separately.

Before you start

  • A Dify workspace with model-provider settings access
  • OpenAI-API-compatible plugin
  • A TokenRex API key

Setup methods

1

Provider setup

  1. 1.

    Open your Dify workspace, or self-host using https://docs.dify.ai/en/self-host/deploy/quick-start/docker-compose.

  2. 2.

    Create a TokenRex API key, then copy an enabled chat model's exact ID from Models & Pricing. The examples use openai/gpt-6-astra; replace it with your selected ID.

  3. 3.

    Install the OpenAI-API-compatible model plugin from Dify's marketplace. Open Settings → Model Provider and add a model under that plugin.

  4. 4.

    Set the model type to LLM, Model Name to openai/gpt-6-astra, API Base URL to https://tokenrex.ai/api/public/v1, API Key to your private TokenRex key, and Completion mode to Chat.

  5. 5.

    If prompted for context size, maximum output or tool-calling capabilities, use the selected model's verified details rather than a guessed value. Save and validate the provider credentials.

  6. 6.

    Create a Chatbot, Agent or Workflow app, choose the configured TokenRex model for its LLM step and send a test message in Preview. Connect and publish channels only after testing.

Configuration
text
Provider: OpenAI-compatible
API base URL: https://tokenrex.ai/api/public/v1
API key: <your TokenRex API key>
Model ID: openai/gpt-6-astra
API mode: Chat Completions
Setup reference

Verify the connection

Send a short message with the selected model, read its response, then confirm the exact model appears in your TokenRex usage. Test streaming and tool use separately before larger agent tasks.