Harness Router: an open-source decision layer that routes agent tool calls with MCTS
Revolutionary_Sir140 · reddit · 2026-09-28
Harness Router is a new open-source, framework-agnostic tool router for agentic systems: instead of letting the LLM blindly pick from a large toolset, it adds a decision layer before execution.
The latest update adds Codex hooks—SessionStart discovers tools once, PreToolUse vets each proposed call (allow if same choice, re-plan if better alternative, fail-open on router failure). For harder decisions it runs Monte Carlo Tree Search over thousands of local simulations.
Goals: fewer wrong tool calls, less redundant tool descriptions in context, lower token usage and reasoning overhead. The author wants to benchmark it on confusing toolsets like getx/fetchx/readx and is soliciting ideas for measuring routing quality beyond success rate.
Related event: Open-Source Harness Router Uses MCTS to Guide Agent Tool Selection(2 posts)→
More from coding & agent
- MemorySec: An Open-Source Security Agent That Remembers Past Incident Remediations — User_0007_123 · 2026-09-28
- Indie dev pays small Twitch streamer to blind-play his game, finds bugs fast — nptacek · 2026-09-28
- EvalSeal v2.2.0 adds tamper-evident eval receipts; LLM judge flipped 5/20 borderline cases — Fit_Fortune953 · 2026-09-28
- Agent-discovered workflow: auto-transcribe YouTube, monitor Substack — yangyi · 2026-09-28
- An AI agent spent the morning drafting outreach for 24 MICCAI 2026 diffusion-synthesis papers — wandedob · 2026-09-28
- How should an MCP server hand arbitrary files like PDFs and Office docs to an LLM agent? — Simple_Newspaper553 · 2026-09-28