llm-alternative

Open reproduction of TypeSafe Jev: a 150M typed decision engine (noul/choice/score in one non-autoregressive pass, calibrated confidence). 0.697 vs Jev's 0.727, 2.5x better calibrated, 4x faster, free. Trains on a Colab T4 in 30 min.

An auditable, hallucination-free alternative to LLMs: 100k facts, sub-millisecond queries, one CPU thread. HRR/VSA substrate with provenance, calibrated confidence, and belief revision. Every benchmark reproducible from the repo.

A deterministic science exploration engine. No LLM, no GPU, no randomness — it restates source records and shows where each subject sits.

Typed questions over your own state, not a chat. A front end for TypeSafe's System One API.

TypeSafe AI is a San Francisco AI lab building System One models — a class of model trained to return typed, calibrated decisions for software instead of generated text. Its flagship and first System One model, Jev, is served by a single HTTP endpoint: POST /v1/systemone takes a `state` (a string, JSON object, or array) plus a map of typed…