Drop a year of chats into one prompt — and ask anything

The idea in one line: you export a year of your chat (or a folder of notes, or your whole journal), ask a plain question — "when did we decide to go to Tbilisi?", "what books did Anya recommend?" — and get an answer drawn strictly from your own messages.

Here's what's fresh. A year ago this wouldn't have been easy. To "talk to your own data," you'd have to build RAG: chop the text into chunks, compute embeddings, store them in a vector database, and search for matching fragments per question. A whole project for one question.

Now the whole year fits in the context window at once. The open model DeepSeek V4-Pro has a million-token context (roughly a 700-page book), and in late May 2026 its 75% price cut was made permanent. A million tokens stopped being expensive exotica.

The surprising takeaway: at personal scale, you often don't need RAG anymore. You just put all the text in the prompt and ask.

What you'll learn

• What a context window is and how much actually fits.
• Why people used to build a vector database for "questions about your data" — and why that's often overkill now.
• How to export a chat or notes into one text file and wire up Q&A in a single call.

A ready starter prompt

A weak request ("here's my chat, tell me something") makes the model invent things. A strong one pins down the source, the format, and a no-making-things-up rule:

"Answer only from this text" and "if it's not there, say so" are the two levers that turn a chatty model into an honest assistant for your own data.

What you'll get

A personal "ask my year" tool. A timeline of your trips. A list of books people recommended. How your mood shifted month to month. All in plain words — no scrolling through thousands of messages by hand.

This is your personal data. Use a model you trust, or run an open one locally — and don't upload someone else's chat without asking.

Start with one export and one question — and you'll feel exactly where the RAG hype ends and "just put it all in the context" begins.

Source: AI Model Release Timeline 2025–2026 — DeepSeek V4-Pro (1M context, MIT)