Ctrl+Z Therapy
CasesQuick answersGuides10-minute checklistHow we testCartoonSend it in

Why does my AI coding tool forget what we built in a long chat?

Ship ItChecked against the official source

Source checked 10 Oct 2026 · Published 10 Oct 2026 · Last reviewed 10 Oct 2026 · By Ctrl+Z Therapy (how we test)

The short answer

The chat is too long. AI coding tools work from a context window: their working memory. In Claude Code, it holds every message, every file Claude reads and every command output. Claude Code's docs say that when it's getting full, Claude may start "forgetting" earlier instructions or making more mistakes. Lovable's say that in very long conversations with a lot of context, instructions may not always be followed consistently. Bolt's say the context window has a limit, so Bolt can forget older messages.

What to do

  1. Write a short notes file the tool reads every time: what your app is and the rules that must not change. Put lasting rules there, not in the chat.
  2. Claude Code: CLAUDE.md. Run /init for a starter file and keep it under 200 lines.
  3. Lovable: Project settings > Knowledge. Bolt: project knowledge (it stays when you clear context). Cursor: a rule (Customize > Rules > Add Rule) set to Always Apply.
  4. New task? Start fresh. Cursor: start a new chat. Claude Code and Bolt: type /clear. In Lovable, keep what matters in Knowledge.

Good to know

  • A notes file helps; it doesn't make forgetting impossible. Claude Code treats CLAUDE.md as context, not enforced configuration, and Lovable says even Knowledge may not always be followed consistently in very long conversations.
  • An old detail matters again? Say it again. Bolt's docs suggest restating it to bring it back into context.
  • No tool gives a number of messages after which it forgets, so we don't either.

Source

Checked on 10 Oct 2026. Prices, limits and settings change, so check the source before you rely on them.

Related