OpenAI Object

An extension for Clickteam Fusion 2.5 that puts the OpenAI Responses API in your event editor.

OpenAI Object lets your Fusion 2.5 game or application talk to OpenAI models directly from events — no DLL wrangling, no external helper process, no blocking your frame. Drop the object into a frame, set your API key, send a prompt, and react to the reply with ordinary Fusion conditions and expressions.

Built on the OpenAI Responses API

The extension speaks the modern Responses API, so you get access to current models and to the features that matter for real applications: instructions, reasoning effort for reasoning-capable models, output token limits, and full control over the request when you need it.

Set a custom base URL if you route requests through your own backend or a compatible proxy.

Asynchronous and non-blocking

Every request runs on a background worker thread. Your frame keeps rendering, your animations keep playing, and your input keeps responding while the model works. Results are handed back safely on the Fusion thread and surface as triggered conditions:

• On response started
• On response completed
• On response failed
• On tool call

You can cancel an in-flight request at any time.

AI responses in your events

Read the generated text with a single expression. You also get the response id, the status, and the complete raw JSON body if you want to dig into anything the higher-level expressions don't expose.

Tool calling driven by Fusion events

Declare function tools with a name, a description, and a JSON schema, then let the model call into logic you implement in Fusion:

• Read the requested tool name and its call id
• Read arguments as strings, numbers, integers, or booleans — including nested fields via dotted paths like target.name
• Handle multiple tool calls in a single turn, each with its own call id

Submit tool results and continue the response

Answer a tool call with Submit tool result and the response continues from exactly where it paused. Results are matched back to the model's call_id, so parallel calls stay correctly paired.

Structured outputs

Supply a JSON schema and the extension builds a strict json_schema response format for you. The model returns JSON matching your shape, and a dedicated expression hands it back unaltered so you can parse it straight into your own data. No more scraping prose for values.

Conversation state that just works

Multi-turn context is handled automatically and is on by default. Sending text appends the user turn; a completed response appends the assistant turn. The next request already knows the conversation.

You stay in control:

• Turn history on or off
• Clear it
• Set a sliding-window limit on stored messages
• Inspect it with a message-count expression and a full history-JSON expression

Usage and token information

Read input and output token counts from the last response, so you can display costs, budget requests, or log usage.

Also included

• Fixed 32x32 editor object with a proper Fusion icon and frame-editor image
• Full Markdown help documentation for every action, condition, and expression
• A ready-to-open example .mfa demonstrating a working chat flow
• Native automated tests behind every release build

Honest scope for v1.0.0

This release does not include streaming (token-by-token output) or image input. Requests are non-blocking, but the response arrives complete rather than incrementally. Conversation history is kept in memory for the lifetime of the running application.

Requirements

• Clickteam Fusion 2.5 (32-bit extension)
• An OpenAI API key — usage is billed to your own OpenAI account
• Windows

Licence

OpenAI Object is a commercial, proprietary extension. You may include it in the applications you build and distribute those applications freely, and you're welcome to make tutorials, videos, and documentation about it. Redistributing the raw extension itself — the .mfx, the source, or the release archive — is not permitted. See the included licence for the full terms.

By LexianDEV. Not affiliated with OpenAI or Clickteam.