SGA-MCTS: Decoupling Planning from Execution via Training-Free Atomic Experience Retrieval

TL;DR

SGA-MCTS introduces a retrieval-based planning framework for LLMs that enables real-time, scalable reasoning by distilling high-quality trajectories into reusable primitives, matching state-of-the-art performance without fine-tuning.

Contribution

It presents a novel non-parametric retrieval approach using MCTS to create reusable symbolic primitives, enabling fast and effective LLM planning without task-specific training.

Findings

Matches GPT-5 performance on complex benchmarks

Enables real-time autonomous planning with frozen models

Amortizes search costs for scalable reasoning

Abstract

LLM-powered systems require complex multi-step decision-making abilities to solve real-world tasks, yet current planning approaches face a trade-off between the high latency of inference-time search and the limited generalization of supervised fine-tuning. To address this limitation, we introduce \textbf{SGA-MCTS}, a framework that casts LLM planning as non-parametric retrieval. Offline, we leverage Monte Carlo Tree Search (MCTS) to explore the solution space and distill high-fidelity trajectories into State-Goal-Action (SGA) atoms. These atoms are de-lexicalized primitives that abstract concrete entities into symbolic slots, preserving reusable causal logic while discarding domain-specific noise. Online, a retrieval-augmented agent employs a hybrid symbolic-semantic mechanism to fetch relevant SGAs and re-ground them into the current context as soft reasoning hints. Empirical results on complex benchmarks demonstrate that this paradigm enables frozen, open-weights models to match the performance of SOTA systems (e.g., GPT-5) without task-specific fine-tuning. By effectively amortizing the heavy computational cost of search, SGA-MCTS achieves System 2 reasoning depth at System 1 inference speeds, rendering autonomous planning both scalable and real-time feasible.