Ai

Claude Mem v12.4.5 Fixes Fresh-Install Memory Persistence Failures

Claude Mem v12.4.5 Fixes Fresh-Install Memory Persistence Failures

Claude Mem v12.4.5 delivers a targeted maintenance update that fixes a fresh-install database migration issue which silently prevented pending message queue claims and observation inserts from reaching memory. The release focuses on restoring reliable persistence behavior for new deployments and adds self-healing protections for already affected databases.

What Changed

The core bug came from SessionStore missing migration 28 column additions when creating fresh pending_messages tables. As a result, newly installed environments were missing the tool_use_id and worker_pid columns, causing queue claim operations and observation writes to fail with “no such column” errors.

Version 12.4.5 fixes this by adding addPendingMessagesToolUseIdAndWorkerPidColumns to SessionStore.ts. The patch mirrors the project’s existing migration safety pattern already used for addObservationSubagentColumns and addObservationsUniqueContentHashIndex.

The update also adds a recovery path for previously broken databases that had advanced to v29 without the required v28 schema columns. On the next worker boot, column-existence guards now allow those installations to self-heal automatically.

In addition, the deduplication DELETE operation and UNIQUE index creation are now wrapped in a transaction, aligning this behavior with the existing v29 mirror precedent and reducing the risk of partial schema repair states.

Why It Matters

This release is important for teams relying on local memory persistence and background worker processing, especially in AI-enabled tooling where silent data loss can undermine trust in the system. Fresh installs are often the first experience developers have with a project, so fixing a migration gap that blocked memory ingestion removes a major reliability issue.

The self-healing behavior also matters operationally. Instead of requiring manual database intervention for installs stuck in an inconsistent schema state, workers can now detect the missing columns and repair the issue during startup. That lowers support overhead and reduces downtime for affected users.

For engineering teams, the transaction wrapping around deduplication and index creation is another meaningful hardening step. It improves migration consistency and reinforces safer database evolution practices in production-like environments.

Official Source: https://github.com/thedotmack/claude-mem/releases/tag/v12.4.5