Snowflake ID Parser

Twitter created the format in 2010 when moving off MySQL auto-increment: with several databases, nobody can be the single authority handing out the next number. The solution was to assemble the identifier from three parts each machine can fill in alone — the clock, its own number and a local counter.

The layout is fixed: one sign bit always zero, 41 bits of milliseconds counted from an epoch the project chooses, 10 bits identifying the machine and 12 bits of sequence. The 12 sequence bits cap it at 4096 identifiers per millisecond per machine, which in practice never binds. It is the 41 time bits that set the scheme's lifespan: about 69 years from the epoch.

The epoch is what varies between implementations, which is why the page lets you choose: Twitter uses 4 November 2010, Discord 1 January 2015. Applying the wrong epoch raises no error — it returns a plausible date that is wrong by years. Worth noting too that sorting by identifier sorts by time, which is half the reason the format exists.