Hacker News
Users are reporting that Claude.ai's auto-compact feature has been failing since January 14, preventing conversations from continuing when the context window fills. This issue was marked as fixed on January 15 but persists, with no official acknowledgment from Anthropic.
AI 生成摘要
用戶報告稱，Claude.ai 的自動壓縮功能自1月14日起即失效，導致在上下文窗口滿載時對話無法繼續。此問題雖於1月15日被標記為已修復，但至今仍存在，且官方尚未承認。
We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Auto-compact doesn't work. Anthropic flagged this as fixed on January 15, but the issue persists. When context window fills up, instead of auto-compacting and continuing the conversation, one of two things happens:
This happens even when the context being fed the chat shouldn't be anywhere near the 200k token limit.
Auto-compact should trigger automatically when approaching context limits, allowing the conversation to continue without manual intervention.
Not sure / Multiple models
Yes, this worked in a previous version
Claude.ai (web/desktop) worked prior to January 14 major outage and January 15 compaction incident. Broken since.
N/A
Other
Windows
Other
Issue was reported previously and marked as resolved January 15. Still broken as of January 17.
Happens within Projects, unsure if it affects non-Project chats.
Note: Submitting here because there's no dedicated bug tracker for Claude.ai. This affects the web and desktop interfaces, not Claude Code specifically. Auto-compact was functional before the January 14 outage and has remained broken despite being flagged as fixed on January 15.