Context Window Management
Audience: professionals who use Claude at work, no coding or API knowledge needed. Verified = checked against Anthropic's documentation or help center on 2026-10-16. Design guidance = widely taught practice, not a documented rule.
How it works (verified, from Anthropic's documentation)
- Claude can only consider a limited amount of text at once: the context window, its "working memory". It includes everything in the conversation: your messages, Claude's replies, pasted documents and files.
- More is not automatically better. As the conversation grows, accuracy and recall can degrade (called context rot), so what is in the window matters as much as how much space remains.
- Chat apps can manage the window on a rolling "first in, first out" basis (older content falls away); the documentation notes that claude.ai can do this.
Signs a chat has gotten too long (design guidance)
- Claude forgets earlier instructions or facts.
- Answers get vaguer or repeat themselves.
- It contradicts something decided earlier.
- You keep re-explaining the same requirement.
Restart, summarise or persist
| Situation | Action |
|---|---|
| New, unrelated task | Start a new chat |
| Long chat with decisions worth keeping | Ask Claude for a summary of decisions, constraints and open items, check it, and start a new chat with the summary pasted in |
| Facts or instructions you need every time | Put them in a Project (instructions and knowledge) so every chat starts with them (D5.2) |
| You want continuity across chats | Use chat search and memory (below) |
| A very large document | Give Claude the part that matters, or ask it to work section by section |
Put the most important instructions at the start and re-state key constraints when a chat is long.
Memory and chat search (verified, from the help center)
- You can ask Claude to search previous conversations; searches use retrieval and appear as tool calls. Claude can also remember context from chats and carry it into new ones.
- Search is limited to all chats outside projects, or within one project's conversations.
- You can turn "Search and reference chats" off in Settings > Memory.
- Incognito chats are temporary and are not pulled into searches. On Team and Enterprise, incognito chats are included in standard data exports and follow the organisation's retention policy.
- Memory is helpful context, not a guarantee; check anything important.
Exam traps
- Keeping one endless chat for unrelated work.
- Starting fresh without carrying over decisions.
- Assuming Claude remembers everything from earlier in a long chat or from past chats.
- Putting sensitive data in memory or projects your organisation does not allow.
References
- Context windows — checked Oct 16, 2026
- Use Claude's chat search and memory — checked Oct 16, 2026
Check your understanding
Claude starts ignoring a rule you gave 80 messages ago. Best fix?Show answerHide answer
Answer
Start a new chat with a summary, or restate the rule (or put it in a Project).Where do facts needed in every chat belong?Show answerHide answer
Answer
In a Project's instructions or knowledge.Does an incognito chat show up in memory search?Show answerHide answer
Answer
No; it is not pulled into searches.