Show HN: Seahaven – Open-source framework for building RL environments

Category: ai-ml

Tags: agent-evals, reinforcement-learning, synthetic-environments, openenv, mcp, python

Score: 7.3/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 7)

Seahaven is a Python framework for building synthetic, stateful worlds to evaluate and train AI agents, with isolated SQLite instances, fixtures, and concurrency support. It stands out by focusing on reproducible, parallel agent runs with change-log-based grading, plus out-of-the-box serving via OpenEnv and MCP.

Target audience: AI/ML researchers and engineers building agent evals or RL pipelines

Repository: https://github.com/Kiln-AI/Seahaven · Python · MIT · 61 stars

View on Hacker News