NOPE Labs

Open & experimental work

Models you can self-host, research artifacts, and projects we ship in the open. Everything here is released as-is — our production APIs live at docs.nope.net.

Projects & tools

Things you can use, self-host, or build on.

Predicate

Experimental

Write the rule in English, get a calibrated score

A promptable semantic scoring function: describe any criterion in plain language and get a reproducible 0–1 probability that a piece of content satisfies it. One small probe, any criterion — no per-criterion training.

Explore

Ocular (Research Build)

Stable

Source-available behavioral embedding classifier

A source-available research build of Ocular — a probe-based classifier for fine-grained behavioral signals in human-AI conversation. PolyForm Noncommercial 1.0.0: noncommercial / research use only (not open source).

System Prompt

Stable

A drop-in safety prompt for any chatbot

An open-source, annotated system prompt for safety-conscious AI assistants. Informed by recognized risk frameworks, false-positive calibrated, and culturally aware — copy it, download it, adapt it. MIT-licensed.

Explore

Research & write-ups

Experiments and findings we’ve published in the open.

Tic Index

Experimental

What language models do when they talk to themselves

A dynamical-systems probe of LLM style: force a model to reply to itself and it collapses into an attractor, its trained stylistic bias self-amplified. We map each model’s "default self", trace the tics to the preference-tuning stage that installs them and the reward models that prefer them, and show one tic is a causally steerable direction. Early findings, methodology-forward.

Explore