Training Data

Last updated: 2026-11-02.

This page explains the "Help improve ytimesy" setting: what it lets ytimesy use, what it never uses, and where things stand today. It is part of our Privacy Policy.

Your choice

  • New accounts see a "Help improve ytimesy" card as the last step of setup. Its switch starts on (off if you are in the EU, EEA, UK or Switzerland); you can change it right there. Your choice (on or off) is saved only when you press Continue, together with the version of this policy shown.
  • Accounts created before the card existed, which never chose, see the same card once in the app, with the same starting position. Continue saves the choice. Closing the card saves nothing, and the setting stays off.
  • Until you press Continue or change it in Settings, the setting is off. A change to this policy never turns it on for you.
  • You can change it at any time in Settings → Privacy. Turning it on there asks you to confirm first.
  • When the setting is on, each project's settings show "Never use this project for training". Use it for client or employer repositories. A project setting can only exclude a project, never include one.

Whether a turn is eligible is decided when the turn starts: the setting is on and the project is not excluded. This applies on every model route.

What can be used

When the setting is on, these parts of chats that ran while it was on can be used. They are what you and your environment contributed:

  • your prompts and messages;
  • your code and repository context: file contents and command output as they were before the agent changed those files;
  • results from your environment: command and test output, exit codes, which files changed (paths and line counts), errors, and pull request outcomes such as merged or closed;
  • feedback you give on a turn (ytimesy has no feedback control yet).

What is never used

Even when the setting is on, these are never training data:

  • anything an AI model wrote, from any provider or route (OpenAI, Anthropic, xAI, OpenRouter, Google, ChatGPT/Codex, Claude Code and Grok subscriptions): replies, reasoning, plans, questions, the commands and tool arguments it chose, web searches it ran, the code and diffs it generated, a subagent's reports, and files the agent changed when they are read back;
  • a local agent session you continued in ytimesy (its transcript holds model replies);
  • message attachments and pull request evidence (screenshots, recordings);
  • your memory repository;
  • credentials stored in ytimesy: model keys, subscription tokens, GitHub tokens, project secrets, MCP sign-ins. They are kept outside chats;
  • chats in a project marked "Never use this project for training";
  • anything from turns that ran while the setting was off.

A secret that appears inside a chat, for example pasted into a message or printed by a command, is part of that chat. ytimesy does not yet remove such secrets automatically. Leave the setting off, or exclude the project, for code with secrets or code you are not allowed to share.

What it is for

Only:

  1. improving ytimesy's agents (prompts, tools, workflows) and building evaluation sets that test them;
  2. training or fine-tuning models that ytimesy builds or commissions to power ytimesy.

We do not sell training data, and we do not give it to model providers or other third parties for their own use.

Where things stand today

There is no training export yet. When the setting is on, eligible turns are marked as eligible in our main database and stay there like any other chat. Nothing has been copied out for training, and no model has been trained on them. Before we copy any eligible data for training, we will update this policy to say how it is de-identified, stored and deleted.

Turning it off

  • Turning the setting off takes effect at once for new turns, and marks every earlier turn of yours as not eligible at the same moment.
  • Excluding a project marks that project's earlier turns as not eligible.
  • Deleting your account deletes your chats and your setting. We keep the history of your setting changes (on, off, when, and the policy version) as proof of consent and withdrawal.

Version 2026-11-02