aiqbench analysis
ToolCompass addresses the critical challenge of tool generalization in LLM agents by optimizing exploration strategies without incurring inference overhead. By structuring tool-call representations into functional distributions, it effectively directs agents toward unseen tools while avoiding wasteful trials on unrelated alternatives. This advancement significantly improves out-of-domain task success rates across multiple reinforcement learning benchmarks.
Toolcompass: Guiding Tool Trialing, Not Suppressing It
arXiv:2609.25678v1 Announce Type: new Abstract: Large language model (LLM) agents must generalize from tools seen during training to unseen tools at deployment. A key challenge is tool trialing, i.e., excessive trials waste the interaction budget, whereas selective trials enable exploration of unfamiliar tools. Existing outcome-based post-training leaves wasteful trials unguided, while turn-level supervision may suppress necessary exploration. We introduce ToolCompass, a post-training framework that guides tool trialing by organizing tool-call representations according to shared functions. Specifically, ToolCompass models each function class as a von Mises--Fisher distribution and jointly reduces intra-function variation across domains and increases inter-function separation. This structure transfers experience from seen tools to functionally similar unseen tools, directing exploration away from unrelated alternatives. ToolCompass requires no ground-truth call traces or unseen-tool access and incurs no inference overhead. Experiments on AppWorld and FTRL show consistent gains across GRPO, RFT, and DMPO. improves AppWorld OOD task success by up to 10.71 percentage points over vanilla post-training and performs best among competitive baselines on both benchmarks.
Abstract:Large language model (LLM) agents must generalize from tools seen during training to unseen tools at deployment. A key challenge is tool trialing, i.e., excessive trials waste the interaction budget, whereas selective trials enable exploration of unfamiliar tools. Existing outcome-based post-training leaves wasteful trials unguided, while turn-level supervision may suppress necessary exploration. We introduce ToolCompass, a post-training framework that guides tool trialing by organizing tool-call representations according to shared functions. Specifically, ToolCompass models each function class as a von Mises--Fisher distribution and jointly reduces intra-function variation across domains and increases inter-function separation. This structure transfers experience from seen tools to functionally similar unseen tools, directing exploration away from unrelated alternatives. ToolCompass requires no ground-truth call traces or unseen-tool access and incurs no inference overhead. Experiments on AppWorld and FTRL show consistent gains across GRPO, RFT, and DMPO. improves AppWorld OOD task success by up to 10.71 percentage points over vanilla post-training and performs best among competitive baselines on both benchmarks.
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2609.25678 [cs.AI]
(or arXiv:2609.25678v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.25678 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 257 words
Related coverage
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · Oct 8, 2026
Don't Judge an LLM Only by Its Activations: Discovering Suppressed Safety Features via Counterfactual Activation Potential
arXiv · Oct 7, 2026
Links #3: 2026/06 Part 1
LessWrong · Jun 15, 2026
Backpropagation-Free Test-Time Adaptation for Lightweight EEG-Based Brain-Computer Interfaces
arXiv · Jul 24, 2026