Quick fixes for the issues that come up first. If your problem is not here, the FAQ and the issue tracker are next.
ConfigError on reopenThe most common cause is a dimension or metric mismatch against the on-disk IDENTITY. Inspect it:
elips info /path/to/dbPass the values from IDENTITY when calling open(), or omit them entirely — existing databases always reopen with their persisted identity. A second cause is a Python callable embedder that was not provided on reopen.
LockConflictAnother process holds the write lock on the database directory. Find it with lsof <db>/LOCK on Linux/macOS. If the previous owner crashed, the OS releases the lock automatically.
Run elips verify to print the recovered prefix and confirm any tail truncation:
elips verify /path/to/dbHNSW graphs are warm-loaded on first use. Subsequent queries hit the resident graph and are sharply faster. If you serve from cold starts often, consider a warm-up query at process start.
Open an issue with the output of elips info, your config, and a minimal reproduction. See also the FAQ.