EN

ContextOS · for those who build with AI

Your AI chats, condensed into a single short context

After weeks of working with Claude, ChatGPT or another assistant, your project's information is scattered across dozens of chats and documents: it repeats itself, contradicts itself, and every new chat either starts from scratch or drags everything along with it. ContextOS reads it all, removes duplicates and outdated material, and gives you a single file to paste at the start of the new chat.

  • Fewer tokens in every chat: paste one short file instead of whole chats and documents. We show you the tokens before and after, counted rather than estimated.
  • No outdated information: when one chat says one thing and a more recent one says another, the file keeps the current version and marks the old one as outdated.
  • Doubts stay doubts: if two sources of equal weight disagree, the file tells the AI to ask you, rather than choosing at random.
  • Your decisions override everything: write down what you have decided and from then on that is what counts.
  • Works with any AI: Claude (Projects or CLAUDE.md), ChatGPT, Gemini, Cursor, Copilot. You can also upload the full export of your Claude or ChatGPT chats.

What it doesn't do: it doesn't write code for you and it doesn't promise that the AI will never make mistakes again. How much you save depends on how much repetition there is in your chats: you can see it measured in the dashboard, project by project.

1. Upload

AI chats, documents, notes, configuration files, your decisions. Paste them in or upload files, including the full export of your chats.

2. Condense

One click: the engine finds duplicates, outdated information and points of disagreement. It takes a few minutes.

3. Paste

Download the context package and put it at the start of the new chat, in the project instructions or in CLAUDE.md.

How to use it

  1. Gather your sources. You can paste a chat by copying it from the assistant's window, or upload the full export: in Claude from Settings → Privacy → Export data; in ChatGPT from Settings → Data controls → Export data. A .zip file arrives by email: inside it there is conversations.json, upload it as it is and each conversation becomes a source with its own date. Also add the project's documents and notes (README, specifications, configuration files).
  2. Say what each source is. Type and date are used to choose the right value when sources disagree: a "my decision" overrides everything, then code and configuration, then documents, then chats; for the same type, the most recent wins. Example: a chat from 2 March says "database: SQLite", one from 20 March says "database: PostgreSQL" → PostgreSQL stays in the package and SQLite is marked as outdated.
  3. Press "Condense". In a few minutes you have the package with the tokens before and after.
  4. Use the package instead of the old chats. Claude: in the instructions or among the files of a Project. Claude Code: in the file CLAUDE.md of the project. ChatGPT: in the instructions of a project or a GPT. Cursor, Copilot and other editors: in the project's rules file. Any other AI: paste it as the first message of a new chat.
  5. Keep it up to date. When you've had some new chats, add them and run it again. If there are any «Da decidere» items, add a source called «Decisione mia» with your choice (for example «hosting: Railway») and run it again: the item will disappear from the open questions.

App menus may change names with updates: if you can't find the option, search for «export data» in the settings.

Your data

Only you can see the sources you upload, from your dashboard with your personal link, and we don't use them for anything else. To analyse them, we send them to the Anthropic (Claude) API, which does not use them to train its models. You can delete them whenever you like with one click; if you cancel, we will delete them ourselves, along with the packages, at the end of your subscription. Please don't upload passwords, keys or other people's personal data: the service doesn't need them.