Release Notes
Fixed
- Semantic-cache keys distinguish tenant and project boundaries. Replacing an entry does not grow the query index.
- Cached searches reuse short result sets for the same or a smaller retrieval limit, retry larger requests, and count only accepted cache hits. Cache entries use the store version read before retrieval.
memory.md and explanation reports are replaced atomically, follow existing or dangling output symlinks, preserve existing Unix mode bits, and check write access to an existing target.
- Session startup preserves invalid scope files, permits automatic scope creation with a wiki-only legacy configuration, respects
.memd-skip for initialized projects, and reads the central health and exposure ledgers.
- Search logging truncates query previews at Unicode character boundaries.
Changed
- Agent instructions make memory lookup optional and keep project state in repository files.
- New
memory.md and explanation files use Unix mode 0600. Atomic replacement requires write access to the destination directory and changes the inode. It guarantees complete-file visibility for readers but does not guarantee crash recovery.
- Session-start JSON omits
auto_scope_recovered_malformed. Invalid scopes are preserved and reported with skipped: invalid_project_scope.
Install memd 1.7.1
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/fmschulz/memd/releases/download/v1.7.1/memd-installer.sh | sh
Download memd 1.7.1