mnestic
GitHub

ULID identifiers

mnestic

Specific to mnestic 0.8.0 (upstream CozoDB #296, shipped in the fork).

A ULID is a 26-character, Crockford-base32 identifier built from a 48-bit millisecond timestamp followed by 80 bits of randomness. Unlike a random UUIDv4, ULIDs sort lexicographically in time order, which makes them ideal keys for append-only memory streams that you scan by recency.

rand_ulid()

Returns a fresh, lexicographically-sortable ULID string.

?[id, created] := id = rand_ulid(), created = now()

Because the timestamp is the high-order component, sorting by the id sorts by creation time:

# most recent memories first, no separate timestamp column needed
?[id, content] := *memory{ id, content }
:order -id
:limit 20

ulid_timestamp(s)

Extracts the embedded Unix-millisecond timestamp from a ULID string.

?[id, ts] := *memory{ id }, ts = ulid_timestamp(id)

Caution

ulid_timestamp rejects malformed or non-canonical ULIDs — wrong length, invalid character, or a leading character greater than 7 — with an error, rather than silently truncating.

Why this matters for memory

Agentic memory is largely append-only and read back by recency. A sortable string key lets you:

  • order recall by time using the key alone (no secondary timestamp index);
  • page through history with :limit / :offset on a single column;
  • keep globally-unique ids without coordinating a counter.

See also