The nightly memory run survives an upgrade, and recall status says which agents to restart after one. Search results are unchanged.
CHAT_SCHEMA string, and the schema is a hash of them all, shown as p- and ten hex digits. The first index run after upgrading re-reads your chats; chats whose files are gone come back from the archive. Later rebuilds happen only when a release changes one of those strings.recall status lists them.recall status names the agents to restart after an upgrade. An agent's recall MCP server keeps the release it started with until the agent restarts. recall status now lists each server that started before recall was last installed into its environment, with the agent that started it (and, for Claude Code, the chat), and --json carries them as outlived. The menu bar shows them under Needs you.changes/<topic>.<section>.md fragment instead of editing CHANGELOG.md, and tools/changelog.py --release compiles them at release. No change for users.no such column: s.root. Now a recall process that an upgrade has outlived leaves the index alone and asks for the agent to be restarted, and a nightly run that finds the chat tables changed under it leaves the chats for the next run and records why. After upgrading, restart agents that were already open.