Koboshi

Import from ChatGPT

Get your words out of ChatGPT.

AI chat is a great teacher and a bad notebook. Keep using it to learn, stop using it to store.

Start free Get the prompt

Do you want to add words while using ChatGPT, or a coding agent like Claude Code? Connect Koboshi to your AI tool and save a word without leaving the chat, or ask it to save every word from the conversation you are in, up to fifty at a time. The prompt further down this page is for conversations that already ended.

The same word, two places

The same lookup, in ChatGPT and in Koboshi.

ChatGPT's answer is good, just buried. Koboshi keeps the one line you needed.

Bad notebook Looking it up in ChatGPT
ChatGPT
Youwhat does “sobremesa” mean?

Great question! Sobremesa is a wonderful and quite charming Spanish word that, interestingly, has no single direct equivalent in English, which is part of what makes it so fun to explore. Let me break it down for you thoroughly.

Etymology

It is formed from two parts: sobre, meaning “over” or “on top of,” and mesa, meaning “table.” So literally it gestures at something like “over-the-table,” which gives you a nice mental image to anchor the meaning.

What it actually means

In everyday use, sobremesa refers to the time spent lingering at the table after a meal, talking and relaxing once the food is finished. It is less an activity than a cherished social ritual.

Cultural context

In Spain and across much of Latin America, the sobremesa can stretch on for an hour or more, especially on weekends or holidays, and is considered an important part of family and social life. It is where conversation, coffee, and sometimes a digestivo happen.

Example sentences
  1. “Nos quedamos de sobremesa hasta las cinco.” (We stayed chatting at the table until five.)
  2. “La sobremesa es mi parte favorita de la comida.”
  3. “Después de comer, siempre hay sobremesa.”
Related words and nuances

You might also encounter tertulia (a more organized social gathering), and it pairs naturally with verbs like charlar and disfrutar. Would you like me to give you more examples, a few synonyms, or a short dialogue using the word in context?

One line you needed, buried in 180 you didn’t, and gone when the chat scrolls.
Good memory The same word in Koboshi
Koboshi
sobremesa so-bre-ME-sa B1ES

Lingering at the table after a meal.

«Nos quedamos de sobremesa.» We lingered after lunch.

Synonymstertulia · charla

flâner flah-NAY B2FR

To wander the city with no aim.

«J’aime flâner en ville.» I love wandering town.

Synonymsse balader · déambuler

木漏れ日 ko-mo-RE-bi C1JA

Sunlight filtering through leaves.

木漏れ日が差す。 Light filters through.

One line each. Scannable, saved, and brought back before you forget.

Your words are not lost. Let's get them out.

Start free

The extraction prompt

Paste this into your chat. Copy the CSV it gives back.

Extraction prompt 
Go through this entire conversation and extract every vocabulary word or phrase that I asked about or that you explained to me.

Rules:
1. Only include words that actually appear in this conversation. Do not add new or related words.
2. One row per word.
3. Output a CSV code block with exactly this header:
word,language,meaning,example
4. Keep each meaning under 12 words. If an example sentence was already used in this chat, reuse it. If not, leave the example field empty.
5. If you are not sure a word was really discussed, leave it out.

If the list gets long, stop before you hit your output limit and write CONTINUE as the last line. I will ask you to continue.

The header matches Koboshi's import format, so the CSV drops straight in.

If the chat is long

Fallbacks, in order.

  1. Continue. “Continue the CSV from where you stopped. Do not repeat rows. Same columns.”
  2. Halves. Run the prompt on the first half, then: “Now the second half. Here is the CSV so far. Add only new words. Do not change existing rows.”
  3. Verify. “Check this CSV against our conversation. Remove any row that does not actually appear. Output the cleaned CSV.” Models can invent words in very long chats. This pass catches it.
  4. Strongest model. If you can pick a model, pick the most capable one. Long chats overflow smaller models first.
  5. Export route. ChatGPT Settings, Data controls, Export data. You get a download link by email. Open the conversation file, paste it in parts into a fresh chat, run the same prompt. A fresh chat means a fresh context window.

Then import

Upload the CSV. The part the chat never did begins.

Upload the CSV, and the reviews start bringing your words back before you forget them.

Questions

Before you run it.

Can the AI hallucinate words that were never in my chat?

It can, in very long chats. The prompt's “only words that actually appear” rule plus the verify pass reduce it a lot; see the steps for long chats.

My chat is months long. Will this work?

Yes. Use halves or the export route.

Does this work with Gemini or Claude chats too?

Yes. The same prompt works in Gemini and Claude.

Where next

New words arrive every day, not just in old chats.

The daily save-and-review rhythm Keep your Duolingo habit, and keep every word you meet beside it.

Your words, out of the chat and into review.

Run the prompt, upload the CSV, and let the reviews do the rest.

Start free