OpenAI API Desk

Connection

API key

Attention: Your API token is sent to OpenAI to process requests. Don’t use keys you aren’t willing to share with a browser app.

Not connected. Enter your key and press INIT.

Model

Show all models 0 models

Ignore inferred limits (send as configured)

Choose a model to see which options it accepts.

Instructions

Instructions sent to the model as developer instructions.

Request parameters

Max output tokens

Maximum number of tokens the model may generate.

Reasoning

Reasoning effort.

Summary

Temperature

Higher values generally make output more random; lower values make it more focused.

Top P

Nucleus sampling. Controls the probability mass considered during token selection.

Usually adjust either Temperature or Top P, rather than both.

Tools

Web search

File search

Code interpreter

Runs Python in an OpenAI-hosted container (container type: auto).

Image generation

Knowledge base

Vector stores hold persistent, searchable files for File search. They are separate from the temporary attachments next to the prompt.

Vector store

New vector store name

Files in the selected store

No vector store selected.

Upload to the selected store

Files are uploaded to your OpenAI account and stay until you delete them.

Operation log

Privacy

Your API key is typed here and kept only in this page's memory. Reloading clears it.
The browser talks directly to api.openai.com. There is no backend, account, or analytics.
The key is always masked in request previews.
Images are sent inline with the request. Documents are uploaded to your OpenAI account and referenced by file ID; Delete uploads removes them.
OpenAI stores Responses by default. This app does not change that setting.

API examples

Educational only. The Request tab always shows what this app actually sends.

Basic request

{
  "model": "gpt-5.6-luna",
  "input": "Explain quantum computing simply."
}

With instructions

{
  "model": "gpt-5.6-luna",
  "instructions": "Answer concisely.",
  "input": "Explain quantum computing."
}

With reasoning

{
  "model": "gpt-5.6-luna",
  "reasoning": {
    "effort": "medium"
  },
  "input": "Solve this problem."
}

With a hosted tool

{
  "model": "gpt-5.6-luna",
  "input": "What changed in the latest release?",
  "tools": [{ "type": "web_search" }]
}

— — input — output — total

Select a model and start a prompt.