Show HN: Jeju – a local-first agent harness with inspectable runs

Category: devtools

Tags: agent-harness, local-first, ai-agents, evaluation, go

Score: 6.5/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 6)

Jeju is a local-first agent harness for defining, running, and inspecting AI agents with explicit boundaries and full trajectory recording. It uniquely combines config-defined behavior, effect-level inspection, and evaluation-guided improvement in a portable bundle, making agent workflows reproducible and auditable.

Target audience: backend devs, devops, data engineers

Repository: https://github.com/cosmtrek/jeju · Go · MIT · 3 stars

View on Hacker News