claude-mem lets users resume coding chats from My Work
The claude-mem project’s 15.19.0 release adds an interface for resuming earlier coding conversations in Claude Code and Codex. Each send starts one continuation with the user’s ordinary permissions. The local interface supports stopping a turn and prevents concurrent turns of its own, while a conversation still running in a terminal presents a remaining collision risk.
Artificial Intelligence··Midday
A new message resumes the original conversation
The claude-mem project, which supplies memory tools for coding assistants, adds conversation continuation in version 15.19.0. A message field in My Work sits with the Claude Code or Codex conversation behind a request. Sending from that field resumes the conversation and places the answer in the interface, keeping messages and replies in sequence. The displayed working directory identifies where execution takes place; if it has vanished, an existing ancestor directory is used.[1]
Failed turns stay stopped
A send creates a single continuation process, and a stop control can end an active turn. Failed work is retained as failed rather than tried again. Startup, reconnecting and timers do not replay old messages. Ordinary user permissions apply, with approval-dependent tool requests rejected and counted in manual mode. An orderly service shutdown stops active turns before exiting.[1]
Local access still allows another local program
Access is restricted to the local computer and other origins are rejected. Without local application programming interface token mode, another local program can still call the routes; the developer recommends that mode on shared machines. My Work blocks its own simultaneous turns in a conversation, but cannot prevent a competing terminal session. The developer warns that concurrent writers can lose messages. A service crash also cannot terminate the external process.[1]