Show HN: Deploy Realistic Personas to Run Conversations in Minutes (github.com)

OneRun is an open-source platform for testing, evaluating, and improving AI agents by simulating realistic conversations at scale. It lets teams spin up diverse personas and adversarial scenarios to run hundreds of multi-turn dialogues per build, automatically surfacing edge cases manual testing misses. Outputs include judge-labeled evaluation conversations for benchmarking, exportable training data (preference pairs, critique-and-revise triples, clean JSONL), and automated QA workflows so you can catch regressions before production—making it useful for RLHF data generation, model evaluation, and continuous integration. Technically, OneRun runs in Docker Compose and relies on a Temporal server for workflow processing; development requires Node.js 22+ and Python 3.12+. The stack includes a FastAPI backend, Next.js frontend, PostgreSQL, and a Python SDK for workflows. Default local endpoints are Frontend: localhost:3000, API: localhost:3001 (with docs), and Temporal UI: localhost:8080. The repo includes scripts to start/stop services, run migrations, and develop locally. For teams building and deploying conversational agents, OneRun streamlines reproducible, scalable simulation, human-in-the-loop evaluation, and dataset export—accelerating robust agent validation and iterative model improvement.