Ditto AI vs OpenAI Record & Replay
Both turn a task you show once on your Mac into something an AI can repeat. Record & Replay lives inside the ChatGPT desktop app and Codex: it drafts a skill and replays it with Computer Use, browser actions and installed plugins. Ditto AI turns the recording into an agent with a real tool for each step, checked with a read-only call, permissions from a fixed table, and installs it into Claude Code or Kiro. Pick Record & Replay if you work in ChatGPT or Codex; pick Ditto if you work in Claude Code or Kiro and want each step on a checked integration.
The short answer
Record & Replay is OpenAI's. In OpenAI's words, it "lets you demonstrate a workflow on your Mac and turn it into a reusable skill". You start it in the ChatGPT desktop app, in Work mode, or in Codex, from Plugins and then Record a skill. After you stop, it "inspects the captured workflow and drafts a skill" that "explains when to use the workflow, what inputs it needs, what steps to follow, and how to verify the result". To replay, you ask a new ChatGPT or Codex chat to use the skill, and it works with "Computer Use, browser actions, and installed plugins" (OpenAI).
Ditto AI is a separate Mac app built around the tools. You record the task once. Ditto writes down the steps, each citing the frames and clicks it came from, then finds a real integration for every step (an MCP server from the official registry or npm, or a headless browser for a public page), checks each one with a single read-only call, and sets permissions from a fixed table. The agent runs in Ditto or installs into Claude Code or Kiro.
The difference that matters day to day: Record & Replay replays the skill with Computer Use, browser actions and whichever plugins you've installed. Ditto puts every step it can on an integration it has found and checked, and shows you before the first run which tool each step will use and what it may do.
Side by side
| Ditto AI | OpenAI Record & Replay | |
|---|---|---|
| Where it runs | A Mac app, macOS 15 or later on Apple silicon | Inside the ChatGPT desktop app (Work) or Codex. "Record & Replay is available on macOS. Computer Use must also be available and enabled" (source) |
| What it captures | One task, while you record it: frames, clicks and, if you talk, a transcript made on the Mac | "ChatGPT or Codex observes the actions and window content needed to learn the workflow"; recording "continues until you stop it" (source) |
| What you get | An agent: steps with their evidence, the inputs that change, a checked tool for every step, and a permission policy | A skill: "when to use the workflow, what inputs it needs, what steps to follow, and how to verify the result" (source) |
| How it repeats the task | Through the tools it found: MCP servers, a headless browser, local tools. A step nothing can do is marked manual | With "Computer Use, browser actions, and installed plugins" (source) |
| Checking tools first | Each tool server is started and makes one real, read-only call before it counts as working | Not stated (source) |
| Permissions | A fixed table per tool; anything unclassifiable is a write. Writes can be held as drafts | Recording starts when you approve the chat's request; admins can switch it off with the computer_use setting. Other controls not stated (source) |
| Which model reads it | The AI provider you connect: your own Anthropic, OpenAI or Amazon Bedrock account, or Ditto AI's | OpenAI's, inside ChatGPT or Codex (source) |
| Where the result runs | Ditto, Claude Code or Kiro CLI. Not Codex yet | ChatGPT or Codex (source) |
| Price | Free; no paid plan yet (pricing) | Not stated on the page; it depends on your ChatGPT or Codex access (source) |
Where Record & Replay is ahead
- Nothing new to install if you already use ChatGPT's Mac app or Codex. It's a menu item, not another app.
- Computer Use reaches apps with no integration. A step with no API behind it can still be done on screen. Ditto marks such a step manual and says so.
- It works in Codex. Ditto doesn't export to Codex yet.
- You can refine the skill in conversation. OpenAI's page says "You can also ask for further refinements."
- It comes from the company that makes the model. One vendor, one account, one bill.
Where Ditto is different
- Each step is tied to its evidence. Every step points at the frames and clicks it came from, and a step with nothing to point at is dropped instead of guessed. Narration can explain a goal, but it can't add a step.
- The tools are named and checked before the first run. You see which server or browser each step uses, and each one has already made a real read-only call. Nothing that writes is exercised to test it.
- Permissions don't come from the model. Access per tool is a lookup in a fixed table, the agent gets only the tools the recording used, and anything that can't be classified counts as a write.
- It isn't tied to one AI company. The recording is read by the provider you choose, including Anthropic and Amazon Bedrock, and the agent runs in Claude Code or Kiro.
- Recordings, agents and credentials stay on your Mac. What leaves is the frames, clicks and transcript of a recording, sent to your chosen provider so it can read the task.
Choose Record & Replay if
- You already work in the ChatGPT desktop app or in Codex.
- The task lives in apps with no integration, and repeating it on screen is fine.
- You want one vendor for the model, the app and the skill.
Choose Ditto AI if
- You work in Claude Code or Kiro and want the agent there, with its tool servers attached.
- You'd rather each step ran on an integration that returns data than on replayed clicks.
- You want to see the evidence, the tools and the exact access before anything runs.
- You want to use your own Anthropic, OpenAI or Bedrock account.
Can you use both?
Yes. They don't conflict on a Mac. If a task already works well as a Record & Replay skill in Codex, keep it there; use Ditto for the ones you want in Claude Code or Kiro, or where you want a checked tool and a fixed policy behind every step.
See the whole pipeline in how Ditto AI works, or read a real export on the templates page.