How it works
How Thread finds your ideas.
Thread doesn't save your chats and it doesn't guess what you meant. It reads what you said, picks out the moments where you had an idea, took a position, made a decision or asked a question, and keeps each one tied to your exact words.
Last updated 21 October 2026. Written from what Thread actually does.
What it reads
Ideas come only from your messages. The AI's replies are shown to Thread's reader so it can follow the conversation, but Thread never takes an idea from something the AI said.
There is one exception, and it is yours: when you plainly accept something the AI proposed (“Yes, the second one”), Thread records the decision you made. It keeps both your words of acceptance and the exact proposal you accepted.
To find ideas, the text of a conversation is read by an AI model. Exactly where it goes and how long it is kept is set out on the privacy page.
What counts as an idea
Only moments that show your thinking, and only the ones that will still matter in a few weeks.
- An idea you start
- “What if billing followed headcount?”
- Something you claim
- “Authority has to be independently verifiable.”
- A question you are trying to settle
- “Do guest collaborators count toward the bill?”
- A decision
- “We're going with yearly terms.”
- A change to an idea you already had
- The same idea, said more sharply than before.
- A conflict with something you said earlier
- Two statements that can't both hold.
- A link between two ideas
- “That's the same problem as the pricing one.”
- An idea you drop
- “Forget the per-workspace option.”
Left out: asking the AI to explain something, instructions like “put that in a table”, thanks and filler, and passing remarks with no real claim in them.
Every idea quotes you
For each moment it keeps, Thread has to point to the exact words you used, copied letter for letter from your message. Then it checks that those words really appear in that message. If they don't, the moment is thrown away. Thread does not put paraphrases in your mouth, and it does not keep an idea it can't trace.
That is why you can always click through from an idea to the conversation and the moment it came from.
From moments to ideas
Thread groups what you said by meaning, not by which conversation or which AI it happened in. Coming back to an idea in a new chat adds to the idea you already have instead of starting a duplicate.
- A question stays with the idea it is about. A question asked right after an idea almost always belongs to it.
- An answer closes the one question it answers, and the answer is shown with it.
- Where an idea stands is your latest position on it, with the history kept. A decision makes it established. Two statements that conflict mark it contested until you resolve it.
- Ideas Thread isn't sure are yours wait in a Maybe list until you keep them, so they don't crowd the real ones.
Grouping is redone from scratch a couple of minutes after new thinking arrives, not decided one message at a time. A wrong early guess is fixed by the next pass instead of piling up.
You have the last word
- Keep an idea, or mark it Not an idea. Each choice teaches Thread what counts as an idea for you.
- Rename an idea, tick or untick a question, pin what matters.
- In the Mac app, merge two ideas that are really one, or split one that is really two.
Your corrections are remembered. Every later regrouping respects them, so a fixed mistake stays fixed.
What it can get wrong
Thread can miss an idea, join two that are different, split one that is whole, or word a thought slightly differently from how you would. It is a reader, not a mind reader. That is the reason every idea links to your exact words and the conversation they came from, and the reason your corrections are kept.