Show HN: Mecatl – Separate what an agent can do from where it runs

Category: ai-ml

Tags: ai-agents, agent-runtime, kubernetes, microvm, golang, grpc, sandboxing

Score: 8.0/10 (Innovation: 8, Technical: 9, Documentation: 8, Utility: 7)

Mecatl is a cloud-native AI agent harness that decouples the agent loop from clients, model providers, and execution environments, exposing pluggable interfaces for state stores, gRPC/HTTP/SSE clients, and optional microVM-isolated execution. It's notable for treating permissions, audit trails, durable sessions, and Kubernetes-native scaling (lease-based session ownership, drain handling) as first-class runtime concerns rather than bolting them onto a vendor stack.

Target audience: backend devs, devops, platform engineers building AI agent infrastructure

Repository: https://github.com/stacklok/mecatl · Go · Apache-2.0 · 288 stars

View on Hacker News