Harness-1: Reinforcement Learning for Search Agents with State-Externalizing Harnesses

2026-06-01

AI Summary

This work addresses the inefficiency and limited generalization of traditional search agents, which couple state management with semantic decision-making, forcing reinforcement learning to jointly optimize high-level reasoning and low-level bookkeeping. To resolve this, the authors propose Harness-1, a 20-billion-parameter retrieval sub-agent trained within a “harness” environment endowed with external state memory. This design delegates working memory tasks—such as maintaining candidate pools and evidence linking—to the environment, allowing the policy to focus exclusively on semantic-level decisions. By innovatively decoupling state management from decision logic and integrating techniques like importance-based passage selection, compressed and deduplicated observations, and budget-aware context rendering, Harness-1 achieves an average top-k recall of 0.730 across eight cross-domain retrieval benchmarks, surpassing the strongest open-source sub-agent by 11.4 percentage points and demonstrating exceptional transfer performance on unseen tasks.

Abstract

Search agents are often trained as policies over growing transcripts: the model must decide how to search while also remembering what it has seen, which evidence is useful, which constraints remain open, and which claims have actually been checked. We argue that this formulation puts too much routine state management inside the policy: reinforcement learning is forced to optimize both semantic search decisions and recoverable bookkeeping that the environment can maintain more reliably. We introduce Harness-1, a 20B search agent (retrieval subagent) trained with reinforcement learning inside a stateful search harness. The harness maintains environment-side working memory, including a candidate pool, an importance-tagged curated set, compact evidence links, verification records, compressed and deduplicated observations, and budget-aware context rendering. The policy retains the semantic decisions: what to search, which documents to keep or discard, what to verify, and when to stop. Across eight retrieval benchmarks spanning web, finance, patents, and multi-hop QA, Harness-1 achieves 0.730 average curated recall, outperforming the next strongest open search subagent by +11.4 points and remaining competitive with much larger frontier-model searchers. Its gains are especially strong on held-out transfer benchmarks, suggesting that reinforcement learning over explicit search state can produce retrieval behaviors that generalize beyond the training domains. Our code is available at https://github.com/pat-jj/harness-1.

Problem

• search agents
• reinforcement learning
• state management
• working memory
• retrieval

Innovation

• state-externalizing harness
• reinforcement learning for search
• working memory offloading
• curated recall
• generalizable retrieval agent