In-The-Flow Agentic System Optimization for Effective Planning and Tool Use

ICLR 2026 Conference Submission16241 Authors

Keywords: Reinforcement Learning, Large Language Models, Agentic Systems, Tool Use, Planning, On-policy Optimization, Sparse Rewards

TL;DR: We introduce AgentFlow, a trainable agentic system, and Flow-GRPO, an on-policy RL algorithm that optimizes the planner "in-the-flow" by broadcasting a final outcome reward to all steps, enabling effective long-horizon planning and tool use.

Abstract: Outcome-driven reinforcement learning has advanced reasoning in large language models (LLMs), but prevailing tool-augmented approaches train a single, monolithic policy that interleaves thoughts and tool calls under full context; this scales poorly with long horizons and diverse tools and generalizes weakly to new scenarios. Agentic systems offer a promising alternative by decomposing work across specialized modules, yet most remain training-free or rely on offline training decoupled from the live dynamics of multi-turn interaction. We introduce AgentFlow, a trainable, *in-the-flow* agentic framework that coordinates four modules (planner, executor, verifier, generator) through an evolving memory and directly optimizes its planner inside the multi-turn loop. To train on-policy in live environments, we propose *Flow-based Group Refined Policy Optimization* (Flow-GRPO), which tackles long-horizon, sparse-reward credit assignment by converting multi-turn optimization into a sequence of tractable single-turn policy updates. It broadcasts a single, verifiable trajectory-level outcome to every turn to align local planner decisions with global success and stabilizes learning with group-normalized advantages. Across ten benchmarks, AgentFlow with a 7B-scale backbone outperforms top-performing baselines with average accuracy gains of 14.9% on search, 14.0% on agentic, 14.5% on mathematical, and 4.1% on scientific tasks, even surpassing larger proprietary models like GPT-4o. Further analyses confirm the benefits of in-the-flow optimization, showing improved planning, enhanced tool-calling reliability, and positive scaling with model size and reasoning turns. Codebase is available at https://anonymous.4open.science/r/agentflow.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 16241