Repository navigation
v0.1.0-alpha.33 — OrbStack support and faster large-project indexing
Pre-releaseCodeMemory now prefers OrbStack on macOS when it is installed and still supports
Docker Desktop. Managed PostgreSQL uses a one-CPU, 1-GiB container limit. The
selected local engine is recorded with the service, so switching contexts cannot
silently replace an initialized database with an empty volume. To intentionally
start a fresh index on another engine, use codememory system setup --fresh-engine;
the old engine volume is left untouched and existing CodeMemory memories are not
copied. Reindex the selected projects after switching.
The indexer fixes Unicode preview truncation that could abort a large publication.
On macOS, the watcher uses a native recursive subscription instead of crawling the
whole tree at startup. TypeScript compiler graphs are released between workspace
groups, and the fallback reconciliation scan runs every five minutes by default.
Database publication now bounds JSON batches by bytes, computes semantic file
digests, and rewrites relationships only for affected files while retaining atomic
publication. Status reads stored index totals instead of recounting large tables;
older index generations continue to use count queries until their next publication.
A disposable-database profiler and focused regressions cover these paths.
Update
On macOS/Linux, run codememory update --version v0.1.0-alpha.33. On Windows,
use codememory.cmd update --version v0.1.0-alpha.33. Reconnect open MCP clients
after updating. The updater preserves project registrations and the selected
engine's managed volume. Database migrations are additive. The first changed or
forced index after updating populates per-file digests and can rewrite existing
relationships once; later publications use those digests for selective writes.
Validation and limits
On macOS, frozen dependency installation, lint, typecheck, build and CLI smoke
check passed. All 236 tests across 43 files passed, as did parser verification,
the upgrade acceptance script and a short disposable-project simulation. The
managed-service acceptance script requires an engine without an existing
CodeMemory service; it remains a hosted Linux release check. The tagged Linux,
Windows and macOS release workflow must pass before GitHub publishes this
prerelease. Compiler memory still scales with project size; the PostgreSQL
container limit does not cap the parser process. Stable-release acceptance
gates remain open.