Harness Router is a decision layer for AI coding agents that makes tool selection faster, cheaper, and more reliable before execution. Use it as a Codex hook to automatically route every tool call, or as a skill when the agent should invoke routing explicitly. Obvious decisions stay on the fast path, ambiguous choices are resolved with Jev, and complex multi-step decisions can escalate to MCTS.

Hey Product Hunt 👋

I built Harness Router to solve a problem I kept seeing in agentic systems: models spending too much time, tokens, and money deciding which tool to call.

The idea is simple: obvious calls take the fast path, ambiguous calls can use Jev, and harder multi-step decisions can use bounded MCTS.

You can integrate it directly as a routing layer, expose it through MCP, or use Codex hooks so tool selection is checked automatically before execution. With SessionStart + PreToolUse, Codex can discover its available tools once and route or re-plan calls through Harness Router transparently.

It’s framework-agnostic and fully open source.

I’m especially interested in feedback from people building coding agents, MCP servers, Codex workflows, or other tool-heavy agent systems. If you try it, I’d love to hear where the routing works well—and where it breaks. 🚀

Today I've added claude hook

A bit of a side note: while working on Harness Router, I also decided to finally publish another project of mine — GoEventBus. 🚀

It’s a high-performance event bus for Go, built around concurrency, multiple backends, and optional intelligent event routing.

If you’re into Go, distributed systems, or event-driven architectures, I’d love to hear what you think:
https://github.com/Protocol-Lattice/GoEventBus