Stop re-explaining yourself to Claude in every new chat

Why every chat starts cold

Claude only knows what's in front of it when the chat starts. Most of what you re-explain falls into three kinds: who you are and how you like answers, the project you're on, and what happened since you last talked about it. Claude has a place for the first two. The third is where the retyping comes from.

1. Profile preferences: who you are

In Claude's settings, the personal preferences apply to every chat. Put things that are always true: your role, your company, the tone and formats you want. Keep it to a few concrete lines. “I'm a freelance designer; answer briefly and ask before writing long drafts” beats a paragraph of adjectives.

2. Projects: one body of work

A Project keeps instructions and files for one client or product, and every chat in it starts with them. Give it:

• Instructions: what the work is, who it's for, what good looks like.
• One short summary of where things stand.
• The few documents you keep pasting from.

The catch is the summary. It's right the day you write it and drifts after that, and Claude can't tell it's out of date.

3. Memory: what came up in chats

With memory on, Claude can draw on earlier conversations and keep a summary of what it learned about you, which you can view and edit in settings. That saves repeating things you've said to Claude before. It doesn't cover the call you had this morning or the thread you read in Slack, because Claude wasn't there. More on how the options compare: how to give Claude memory of your work.

4. What happened outside Claude

This is most of what you re-explain: what was said on calls, what you decided in other apps, what's open with whom. Typing it in each time is the problem you're trying to stop, so it needs to be written down as it happens, somewhere Claude can read.

That's what Cachejay does. On your Mac it keeps the text of what you read and write, transcribes your calls without a bot, and keeps a page per project with what was decided and what's still open. Claude reads it through a connector, so you ask “where did we leave Acme?” instead of explaining it.

Check it worked

Open a fresh chat and ask about something you know the answer to: “What did we decide about the launch date on last week's call with Acme? Give the date.” If Claude answers with the right decision and where it came from, you can stop explaining it. If it guesses, ask it to search first, and check that the call was recorded.

Using ChatGPT too? The same layers exist there: how to stop pasting context into ChatGPT. Why this keeps happening at all: why you keep re-explaining yourself to AI.