LIVE SEMANTIC MEMORY LOOP
Ask once. Rephrase it.
Watch MIRA remember.
The first wording retrieves evidence from a separate public-domain book corpus and generates a grounded answer. A second wording that reaches the same semantic Anchor reuses the stored answer and evidence with zero additional LLM calls.
Checking the local runtime
Is the research Mac awake?
The site is checking the bounded connection to the local semantic-memory prototype.
Two wordings. One answer experience.
Choose one question in a pair, wait for the grounded answer, then choose its paraphrase. The second path should fold into the same Anchor and skip generation.
Run the first wording in a pair. Then run its rephrase and compare the route, model calls, and time.
- ROUTE
- —
- LLM CALLS
- —
- SEMANTIC ANCHOR
- —
- QUESTION MATCH
- —
Generation becomes reusable experience.
- ObserveThe first question is embedded and compared with the isolated English book corpus.
- AnswerRetrieved passages ground one LLM generation, then travel with the answer.
- FoldA paraphrase converging on the same Anchor joins the existing answer experience.
- ReuseThe second answer returns from memory with the same evidence and zero LLM calls.
Demo corpus: Alice’s Adventures in Wonderland, Project Gutenberg eBook #11. This public corpus and its temporary session memory are isolated from MIRA’s private datasets.