Skip to contentATP Token
Integrations

Platforms and workflow tools

Tools where you assemble AI through an interface instead of writing code — app builders, workflow automation — talk to the Gateway the same way an SDK does. Point the tool's OpenAI-compatible provider at the Gateway base URL and every model your project allows becomes available inside it, on one key and one bill.

This page is also plain Markdown

Add .md to any docs URL to read the page as plain text: https://atptoken.ai/docs/platforms.md. Useful when you want to hand the whole page to an assistant.

Before you start

Three values, the same everywhere:

  • Base URL — https://api.atptoken.ai/v1. Include /v1, do not append /chat/completions.
  • API Key — the atp-… project key.
  • Model — an id from GET /v1/models.

Where the tools differ is whether they can read your model list for you:

ToolWhere the settings liveModel list
DifySettings → Model Providersadded one at a time, by hand
n8nthe OpenAI credential's Base URL fieldread from GET /v1/models

If you are wiring up a coding agent instead — Claude Code, Codex, Cline — see Coding agents.

Dify

Install the OpenAI-API-compatible model provider from the marketplace, then use Add Model on its card. The fields that matter:

FieldValue
Model Namethe model id from GET /v1/models
API Keyyour atp-… key
API Base URLhttps://api.atptoken.ai/v1
Model context sizethe model's real context window
Upper bound for max tokensthe model's real output limit
Function Call TypeTool Call if you want agents to use tools
Vision SupportSupport for models that accept images

Two defaults cause most of the confusion. Model context size defaults to 4096, which silently truncates anything longer, so set it per model. And Function Call Type defaults to no_call, which means Dify never sends a tools array — an agent built on that model will quietly refuse to call tools until you change it.

Dify does not read GET /v1/models, so each model is a separate Add Model entry. Call the endpoint first and work from the list it returns.

The setup is identical on Dify Cloud and self-hosted.

n8n

n8n uses its built-in OpenAI credential. Create one, then under Add option set Base URL to https://api.atptoken.ai/v1 and paste the atp-… key into API Key.

The model dropdown then reads GET /v1/models through that base URL, so your project's models appear without any manual list.

To use it, add an OpenAI Chat Model sub-node to an AI Agent or Basic LLM Chain node and select the credential.

Use the AI Agent node, not the OpenAI node's Message a model

The OpenAI node's own Message a model action has an open bug with custom base URLs: the credential test passes but the request returns 404 at runtime. AI Agent plus an OpenAI Chat Model sub-node is the path that works.

Two more notes. The Base URL field lives on the credential, not on the node — it was hidden from the node in a later version, which is why it can look missing. And n8n's own docs do not list the field; it exists in the credential UI under Add option.

The credential behaves the same on n8n Cloud and self-hosted.

Troubleshooting

  • 403 — the model is not enabled for the key's project. Open the project's Resources in the Console and enable it.
  • 404 — /v1 is in the wrong place, or the tool appended /chat/completions to a base URL that already ended in it.
  • The model picker is empty — Dify never lists models for you; add each one by hand. In n8n, an empty list usually means the key or base URL is wrong, since the dropdown is populated from GET /v1/models.
  • Tools are never called — in Dify, Function Call Type is still no_call.
Next steps
← PreviousGoogle GenAI SDK