🔬 Open-source application for running randomised controlled trials (RCT) based human– social AI-agent interaction experiments 🔬
| Repository | What it is | Status |
|---|---|---|
| humaipsy | Cross-platform desktop application (Windows, Linux, macOS) for running multi-condition (e.g. Cooperative vs Argumentative) chat-based experiments with a local or remote Ollama model, with a researcher-configurable number of trials per condition. | |
| Every interaction trial is recorded to a CSV, a mirrored JSONL audit-trail ledger and a local SQLite chat log, and a read-only browser dashboard can show a live view on a secondary screen. | Active · latest release |
Contributions are welcome from people who run experiments, and from people who write software for people who run experiments.
- Questions, methods, ideas — open an issue.
- Building on the engine — start with ARCHITECTURE.md and the contributor notes; the engine has no UI dependency and is meant to be reused.
- Running a study — releases carry an archive per platform plus a
SHA256SUMS.txtmanifest. - Handling participant data — read SECURITY.md before collecting any.
Human–AI interaction as a measurable phenomenon.