100 minigames. One life.
Every level is a different game, and you only get one line of rules. One mistake and you're back to level 1.
How far can you get?
Play free in your browser:
Show more
116,000 stars vs. one binary.
claude-mem needs Node, Bun, uv and Chroma, has a second AI write your memory, and turns analytics on by default.
palace-rs is one binary and one SQLite file. It stores your memory verbatim, uses zero tokens, has no telemetry, and scores 98.7% R@5 on LongMemEval.
Every claim is quoted from the repos. Go check them.
Show more
claude-mem has 116,434 GitHub stars. palace-rs has 10.
I compared what each one actually ships, using only their own READMEs and source:
claude-mem: Node + Bun + uv/Python + Chroma + an HTTP worker. An LLM (Claude Haiku by default) writes your memories. Telemetry on by default. No published retrieval benchmark.
palace-rs: one Rust binary, one SQLite file. Stores what happened verbatim, no LLM, zero tokens to write. No telemetry. LongMemEval R@5 0.987, reproducible.
Stars tell you what people clicked. Not what you have to run.
Full fact-by-fact breakdown below 👇
Show more
