FlowCompile performs compile-time design space exploration on structured LLM workflows to produce reusable high-quality configuration sets that outperform routing baselines with up to 6.4x speedup.
hub Canonical reference
Multi-agent Architecture Search via Agentic Supernet
Canonical reference. 100% of citing Pith papers cite this work as background.
23
Pith papers citing it
Background
100% of classified citations
abstract
Large Language Model (LLM)-empowered multi-agent systems extend the cognitive boundaries of individual agents through disciplined collaboration and interaction, while constructing these systems often requires labor-intensive manual designs. Despite the availability of methods to automate the design of agentic workflows, they typically seek to identify a static, complex, one-size-fits-all system, which, however, fails to dynamically allocate inference resources based on the difficulty and domain of each query. To address this challenge, we shift away from the pursuit of a monolithic agentic system, instead optimizing the \textbf{agentic supernet}, a probabilistic and continuous distribution of agentic architectures. We introduce MaAS, an automated framework that samples query-dependent agentic systems from the supernet, delivering high-quality solutions and tailored resource allocation (\textit{e.g.}, LLM calls, tool calls, token cost). Comprehensive evaluation across six benchmarks demonstrates that MaAS \textbf{(I)} requires only $6\sim45\%$ of the inference costs of existing handcrafted or automated multi-agent systems, \textbf{(II)} surpasses them by $0.54\%\sim11.82\%$, and \textbf{(III)} enjoys superior cross-dataset and cross-LLM-backbone transferability.
hub tools
citation-role summary
background 5
citation-polarity summary
roles
background 5polarities
background 5representative citing papers
HAM³ achieves up to 78.3% attack success rate on the GQA benchmark by hierarchically attacking perception, communication, and reasoning layers in multi-modal multi-agent systems.
TacoMAS performs test-time co-evolution of agent capabilities and communication topology in LLM multi-agent systems via fast capability updates and slow meta-LLM topology edits, delivering 13.3% average gains over strong baselines on four benchmarks.
AgentFlow uses a typed graph DSL covering roles, prompts, tools, topology and protocol plus a runtime-signal feedback loop to optimize multi-agent harnesses, reaching 84.3% on TerminalBench-2 and discovering ten new zero-days in Chrome including two critical sandbox escapes.
MemEvolve jointly evolves agent experiential knowledge and memory architectures via a modular codebase, delivering up to 17% gains on agent benchmarks with cross-task and cross-model generalization.
Method-seeded agent architecture search yields confirmed or directional success-rate gains on four embodied executors, while exposing rollout noise, local basins, and partial credit assignment as hard constraints.
Activation-space divergence detects and corrects compromised LLM agents in multi-agent systems without interaction graphs or synchronized rounds, outperforming graph baselines especially under async stealthy attacks.
A new benchmark study finds that prompt optimization can deliver significant gains in multi-agent LLM systems but its effectiveness varies strongly with task, workflow, communication protocol, and team size.
VTOS jointly searches solution and observer programs to adaptively orchestrate vision tools, outperforming static pipelines on dense object counting and zero-shot plant disease segmentation.
Automatically generated multi-agent systems underperform CoT-SC on benchmarks and a new diagnostic dataset, exposing architectural bloat that fails to deliver functional utility.
StepFinder turns execution logs into temporal semantic sequences via LLMs then uses temporal modeling plus attention to attribute failures to specific steps more accurately and 79% faster than direct LLM methods on the Who&When benchmark.
FALAT improves failure attribution in LLM agent trajectories via dependency-guided search, achieving 46.0% step-level accuracy on algorithm-generated and 29.1% on hand-crafted trajectories in the Who&When benchmark.
AgentCo-op retrieves and assembles existing agents and tools into interoperable workflows for open-world scientific tasks, showing effectiveness in genomics case studies and competitive benchmark results with lower costs.
LEMON trains an LLM orchestrator with counterfactual-augmented GRPO to produce deployable multi-agent specifications that reach state-of-the-art results on six reasoning and coding benchmarks.
EvoMAS trains a workflow adapter with policy gradients to dynamically instantiate stage-specific multi-agent workflows from a fixed agent pool, using explicit task-state construction and terminal success signals, and outperforms static baselines on GAIA, HLE, and DeepResearcher.
An ensemble-based information-theoretic active learning method using ensemble Kalman inversion selects valuable tasks to optimize communication structures in LLM multi-agent systems more reliably than random sampling under limited training budgets.
SBD is a bilevel optimization framework that learns context-dependent safety weights for runtime task delegation in hierarchical multi-agent systems, with continuous authority transfer alpha and theoretical guarantees on safety monotonicity, policy convergence, and accountability propagation.
Evo-Memory is a new streaming benchmark and evaluation framework for self-evolving memory in LLM agents, unifying over ten memory modules and introducing the ReMem pipeline for continual improvement on multi-turn and reasoning datasets.
GenoMAS deploys six specialized LLM agents with guided planning to preprocess transcriptomic data and identify genes, reaching 89.13% composite similarity and 60.48% F1 on the GenoTEX benchmark while outperforming prior methods.
SkillDisCo distills reusable PFSM subgraphs from successful agent traces and compiles them into callable procedural skills, improving success rates and reducing turns on ALFWorld and WebArena.
HarnessFix diagnoses harness flaws from agent traces via HTIR, maps them to repair operators, and improves benchmark performance by 6.3-18.4% over baselines.
The paper delivers the first systematic review of self-evolving agents, structured around what components evolve, when adaptation occurs, and how it is implemented.
A survey unifying the operator-theoretic, probabilistic, and optimization-based mathematical structures underlying modern reinforcement learning algorithms.
citing papers explorer
-
FlowCompile: An Optimizing Compiler for Structured LLM Workflows
FlowCompile performs compile-time design space exploration on structured LLM workflows to produce reusable high-quality configuration sets that outperform routing baselines with up to 6.4x speedup.
-
Hierarchical Attacks for Multi-Modal Multi-Agent Reasoning
HAM³ achieves up to 78.3% attack success rate on the GQA benchmark by hierarchically attacking perception, communication, and reasoning layers in multi-modal multi-agent systems.
-
TacoMAS: Test-Time Co-Evolution of Topology and Capability in LLM-based Multi-Agent Systems
TacoMAS performs test-time co-evolution of agent capabilities and communication topology in LLM multi-agent systems via fast capability updates and slow meta-LLM topology edits, delivering 13.3% average gains over strong baselines on four benchmarks.
-
Synthesizing Multi-Agent Harnesses for Vulnerability Discovery
AgentFlow uses a typed graph DSL covering roles, prompts, tools, topology and protocol plus a runtime-signal feedback loop to optimize multi-agent harnesses, reaching 84.3% on TerminalBench-2 and discovering ten new zero-days in Chrome including two critical sandbox escapes.
-
MemEvolve: Meta-Evolution of Agent Memory Systems
MemEvolve jointly evolves agent experiential knowledge and memory architectures via a modular codebase, delivering up to 17% gains on agent benchmarks with cross-task and cross-model generalization.
-
Automating the Design of Embodied Agent Architectures
Method-seeded agent architecture search yields confirmed or directional success-rate gains on four embodied executors, while exposing rollout noise, local basins, and partial credit assignment as hard constraints.
-
When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems
Activation-space divergence detects and corrects compromised LLM agents in multi-agent systems without interaction graphs or synchronized rounds, outperforming graph baselines especially under async stealthy attacks.
-
MAS-PromptBench: When Does Prompt Optimization Improve Multi-Agent LLM Systems?
A new benchmark study finds that prompt optimization can deliver significant gains in multi-agent LLM systems but its effectiveness varies strongly with task, workflow, communication protocol, and team size.
-
VTOS: Learning to Orchestrate Vision Tools by Co-Searching Solutions and Observers
VTOS jointly searches solution and observer programs to adaptively orchestrate vision tools, outperforming static pipelines on dense object counting and zero-shot plant disease segmentation.
-
The Illusion of Multi-Agent Advantage
Automatically generated multi-agent systems underperform CoT-SC on benchmarks and a new diagnostic dataset, exposing architectural bloat that fails to deliver functional utility.
-
StepFinder: A Temporal Semantic Framework for Failure Attribution in Multi-Agent Systems
StepFinder turns execution logs into temporal semantic sequences via LLMs then uses temporal modeling plus attention to attribute failures to specific steps more accurately and 79% faster than direct LLM methods on the Who&When benchmark.
-
FALAT: Tracing Failures in LLM Agent Trajectories via Dependency-Guided Search
FALAT improves failure attribution in LLM agent trajectories via dependency-guided search, achieving 46.0% step-level accuracy on algorithm-generated and 29.1% on hand-crafted trajectories in the Who&When benchmark.
-
AgentCo-op: Retrieval-Based Synthesis of Interoperable Multi-Agent Workflows
AgentCo-op retrieves and assembles existing agents and tools into interoperable workflows for open-world scientific tasks, showing effectiveness in genomics case studies and competitive benchmark results with lower costs.
-
LEMON: Learning Executable Multi-Agent Orchestration via Counterfactual Reinforcement Learning
LEMON trains an LLM orchestrator with counterfactual-augmented GRPO to produce deployable multi-agent specifications that reach state-of-the-art results on six reasoning and coding benchmarks.
-
EvoMAS: Learning Execution-Time Workflows for Multi-Agent Systems
EvoMAS trains a workflow adapter with policy gradients to dynamically instantiate stage-specific multi-agent workflows from a fixed agent pool, using explicit task-state construction and terminal success signals, and outperforms static baselines on GAIA, HLE, and DeepResearcher.
-
Active Learning for Communication Structure Optimization in LLM-Based Multi-Agent Systems
An ensemble-based information-theoretic active learning method using ensemble Kalman inversion selects valuable tasks to optimize communication structures in LLM multi-agent systems more reliably than random sampling under limited training budgets.
-
Safe Bilevel Delegation (SBD): A Formal Framework for Runtime Delegation Safety in Multi-Agent Systems
SBD is a bilevel optimization framework that learns context-dependent safety weights for runtime task delegation in hierarchical multi-agent systems, with continuous authority transfer alpha and theoretical guarantees on safety monotonicity, policy convergence, and accountability propagation.
-
Evo-Memory: Benchmarking LLM Agent Test-time Learning with Self-Evolving Memory
Evo-Memory is a new streaming benchmark and evaluation framework for self-evolving memory in LLM agents, unifying over ten memory modules and introducing the ReMem pipeline for continual improvement on multi-turn and reasoning datasets.
-
GenoMAS: A Multi-Agent Framework for Scientific Discovery via Code-Driven Gene Expression Analysis
GenoMAS deploys six specialized LLM agents with guided planning to preprocess transcriptomic data and identify genes, reaching 89.13% composite similarity and 60.48% F1 on the GenoTEX benchmark while outperforming prior methods.
-
SKILL-DISCO: Distilling and Compiling Agent Traces into Reusable Procedural Skills
SkillDisCo distills reusable PFSM subgraphs from successful agent traces and compiles them into callable procedural skills, improving success rates and reducing turns on ALFWorld and WebArena.
-
From Failed Trajectories to Reliable LLM Agents: Diagnosing and Repairing Harness Flaws
HarnessFix diagnoses harness flaws from agent traces via HTIR, maps them to repair operators, and improves benchmark performance by 6.3-18.4% over baselines.
-
A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence
The paper delivers the first systematic review of self-evolving agents, structured around what components evolve, when adaptation occurs, and how it is implemented.
-
Mathematical methods of reinforcement learning
A survey unifying the operator-theoretic, probabilistic, and optimization-based mathematical structures underlying modern reinforcement learning algorithms.