- Three local-first research tools are now on PyPI—built for work that should remain private, inspectable, and reproducible. 1. FrontierTrials: blind AI product comparisons 2. BenchLineage: EE experiment provenance 3. ReproWeave: replication triage Details ↓
- FrontierTrials v0.3 is live. A private, no-API lab for deciding which AI subscription to keep: paste exact outputs from ChatGPT, Claude, Gemini, Kimi or GLM, review anonymous pairs, and build a browser-local history. Try it: caoshurong.github.io/frontiertrials…
- Most AI evaluation tools assume an API. I use subscription web apps, so I built FrontierTrials: a local-first toolkit for blind human review of manually captured AI outputs. No API keys and no hosted judge.
