Conversations and blind sessions
Keep ordinary Conversations separate from controlled RV Sessions while preserving the boundary between Viewer evidence and Reveal information.
LOCAL-FIRST · BLIND · REPRODUCIBLE
AI RV Harness combines blind RV sessions, ordinary Conversations, structured Training, controlled Research, separate AI roles, target management, evidence-preserving Reveal flows, and reproducible evaluation tools.
Latest public release: v0.7.14 Branch: main
WHAT IT IS
Keep ordinary Conversations separate from controlled RV Sessions while preserving the boundary between Viewer evidence and Reveal information.
Use bundled and user-provided targets, resumable checkpoints, Viewer self-review, Viewer Notes, and optional judging.
Build locked, blinded studies with reproducible designs, frozen evidence, judging, unblinding, and experiment statistics.
Keep AI Monitor history, AI identities, Viewer Notes, and versioned Viewer Learning resources in distinct profile-wide tools.
The desktop application stores project state locally with SQLite while remote providers are used only when selected.
Viewer, Monitor, Judge, Research, and post-Reveal material are handled through explicit workflow boundaries rather than one blended chat.
CURRENT PUBLIC UI
These screenshots show the current AI RV Harness interface for the v0.7.14 public release. The gallery reflects the current navigation and public project presentation.






RESEARCH PURPOSE
AI RV Harness provides a structured environment for testing individual AI models in Remote Viewing tasks, comparing their capabilities, and evaluating how model and generation parameters influence performance.
PROJECT RESOURCES