DoneThat vs OpenAI Chronicle

The short answer

Chronicle is a feature, not a product. It lets Codex read your recent screen content so it stops asking you to restate your context in every prompt — an opt-in research preview inside the Codex app on macOS, available to ChatGPT Pro subscribers.

It is aimed at making a coding agent better at coding. DoneThat is aimed at telling you where your week went and coaching you on it. Both read your screen; that is where the resemblance ends.

What Chronicle solves

A real and specific irritation: an assistant that helped you an hour ago has no idea what you are working on now. You re-explain the project, the stack, the file you are in, the thing you already ruled out.

Chronicle removes that by giving Codex recent screen context, so it can interpret references and pick up on the tools and workflows you actually use. Inside Codex, that is a genuine improvement, and the integration is tight in a way a third-party tool cannot match.

How the data is handled

Worth stating precisely, because it is more careful than the headlines suggested:

• Screen captures are stored temporarily on your own computer.
• Selected frames are processed through an ephemeral Codex session on OpenAI's servers to generate structured memories.
• Captures older than six hours are deleted automatically while the feature is active.
• It is opt-in, as a research preview.

That is a short-retention, selective design rather than a permanent archive.

DoneThat's approach differs: screenshots are never stored at all — processed in real time, immediately deleted, only the derived structured activity persists. Excluded apps are redacted on your machine before anything is sent. With your own Gemini or OpenAI-compatible model, raw activity never reaches our servers. The desktop app is open source.

Two risks OpenAI flags itself

Credit for naming them:

Rate limits. Chronicle may consume them quickly, so it competes with everything else you use your ChatGPT Pro allowance for. That is an indirect cost that does not appear on a price list.

Prompt injection. Exposure increases because content on your screen becomes potential input to an agent. That is a serious consideration and not a theoretical one — anything visible on your display, including text you did not write, is in scope.

DoneThat's capture feeds a reconstruction rather than an agent with tool access, which is a narrower blast radius, but it is not immune to the same class of problem and we are not going to claim it is.

Different purposes

Chronicle's question: what is this developer working on right now, so Codex can help?

DoneThat's question: where did this week go, and what should change?

Chronicle does not total your time, does not know you have a goal, and will not tell you that your stated priority received two hours. Don does, and reaches out during the work — nudging you onto goals, blocking calendar time for them, flagging low-value work while you can still fix it, checking your mood, reminding you of meetings, suggesting breaks.

There is also a scope difference worth noting: Chronicle's memories live inside Codex. DoneThat exposes reconstructed work history through an MCP server and API, so any MCP-capable assistant — including Codex — can query it.

Availability

macOS only, inside the Codex app, requiring a ChatGPT Pro subscription. There is no separate charge and no free path. As a research preview, expect it to change.

DoneThat runs on Windows, macOS and Linux with a 14-day free trial; current prices are on the DoneThat download page.

Where Chronicle is the better choice

• You already pay for ChatGPT Pro and use Codex daily. It is included and it is right there.
• Your problem is re-explaining context to a coding agent.
• You are on macOS and work primarily in Codex.
• You want short retention. Six-hour automatic deletion is aggressive in a good way.
• You want nothing extra installed.

Where DoneThat is the better choice

• You want to know where your time went. Chronicle does not measure it.
• You use Windows or Linux.
• You want coaching against goals, not better agent context.
• You want more than one assistant to use your history. MCP exposes it broadly rather than to one app.
• You do not want a coding agent reading your screen with the prompt-injection exposure that brings.
• Your team wants shared signal without surveillance. Summary-level dashboards, relative trends rather than per-person hours, visibility set per summary, optional mood tracking.

How we compared these

Statements about Chronicle come from published reporting on its launch, read on 11 August 2026 and linked below, since it is a preview feature rather than a product with its own documentation page. Its stated risks are OpenAI's own, reported as such.

Claims about DoneThat are first-hand: we build it. Chronicle is included with a subscription many developers already hold, its six-hour deletion window is stricter than most retention policies, and inside Codex its integration is better than anything external. The narrow claim we would defend is that context for an agent and a record of your work are different things.

Questions people ask about OpenAI Chronicle

What is OpenAI Chronicle?

An opt-in research preview in the Codex app on macOS that captures recent screen content to build memories, so Codex can interpret references and pick up on your tools and workflows without you restating context in every prompt.

Does Chronicle store screenshots of my screen?

Temporarily. Captures are held on your own computer, selected frames are processed through an ephemeral Codex session on OpenAI's servers to generate structured memories, and captures older than six hours are deleted automatically while the feature is active.

What are the risks OpenAI itself flags?

Two: it may quickly consume your rate limits, and it increases exposure to prompt injection attacks from content on your screen. The second is worth taking seriously — anything visible on your display becomes potential input to an agent.

Chronicle vs DoneThat?

Chronicle makes a coding agent better at coding by feeding it your screen context. DoneThat reconstructs your working day, coaches you against goals, and exposes that history to any MCP-capable assistant. Chronicle is a feature inside one tool; DoneThat is a record of your work you can point several tools at.