terraphim MegaHAL
A 1998 Loebner Prize chatterbot, ported to Rust and WebAssembly. Second-order Markov chains, deterministic seeded replies, zero LLM, zero network calls -- everything runs in your browser.
Original MegaHAL by Jason Hutchens (Unlicense). Rust port:
terraphim_megahal.
Teach MegaHAL (one sentence per line)
Brain Lab -- watch training change the text
Both brains share the same seed, start identical, and receive the same input. Teach Brain B below, then ask both: Brain A keeps answering from its original corpus while Brain B's replies absorb what it learned.
Brain A default personality
-- msBrain B starts identical -- train it
-- msWatch me learn
Type a nonsense word, teach MegaHAL three sentences about it, then ask about it. The counters below are the brain growing in real time, and the words you taught light up inside its replies.