What it is
A Pi coding harness where any LLM writes code while TypeSafe Jev supplies typed reflexes for policy, loop detection, and edit review. Aimed at developers who want calibrated S1 judgments without handing isolation to a model. Explicitly not a sandbox.
How it uses Jev
Per reflex, Bicameral builds redacted state plus a pack of Choice/Score/Noul questions and sends them to Jev, which returns probabilities. Pure decide* functions turn those probabilities, via bicameral.yaml thresholds, into allow, confirm, block, warn, or steer for the Pi agent. Jev never generates text.
Primitives:choicescorenoul
Technique worth stealing
Prefetch the S1 judgment on message_update so Gate's answer is ready before tool_call.
Try it
pnpm install, then pi install "$(pwd)/packages/pi-bicameral"; set TYPESAFE_API_KEY for live Jev.
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.94
- Shows a System One patternyes0.85
- Handles uncertaintyno0.18
- Measuredno0.04
- Runnableyes0.88
- Worth recommendingunclear0.47
- Model replicano0.04
- Problem scopescore on a 0–2 scale1.38
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.