Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
8
8
section
stringclasses
11 values
subcategory
stringlengths
7
43
rtype
stringclasses
6 values
title
stringlengths
10
131
url
stringlengths
23
139
venue
stringlengths
3
54
year
int64
1.98k
2.03k
authors
stringlengths
4
183
description
stringlengths
99
269
why
stringlengths
77
200
evidence
stringclasses
7 values
layer
stringclasses
9 values
age-0001
Start Here
Architecture guide
Blog
A practical guide to building agents
https://openai.com/business/guides-and-resources/a-practical-guide-to-building-ai-agents/
OpenAI Guides & Resources
2,025
OpenAI
Explains when agents are appropriate, how to define their tools and instructions, and how to move from one agent to manager and handoff-based multi-agent patterns.
A useful first check on whether distinct roles justify the coordination cost of a graph.
Practitioner analysis
Roles
age-0002
Start Here
Architecture guide
Blog
Building effective agents
https://www.anthropic.com/engineering/building-effective-agents
Anthropic Engineering
2,024
Erik Schluntz; Barry Zhang
Distinguishes workflows from autonomous agents and presents routing, parallelization, orchestrator-worker, and evaluator-optimizer patterns.
Provides a compact topology vocabulary and a strong case for adding coordination only when the task demands it.
Practitioner analysis
Topology
age-0003
Research Foundations
Agent foundations
Paper
Agent-Oriented Programming
https://doi.org/10.1016/0004-3702(93)90034-9
Artificial Intelligence
1,993
Yoav Shoham
Defines a programming paradigm in which agents are first-class components described through mental state and governed by explicit interaction rules.
Establishes the intellectual lineage for treating an agent role as a programmable organizational unit.
Peer-reviewed research
Roles
age-0004
Research Foundations
Agent foundations
Paper
Intelligent Agents: Theory and Practice
https://doi.org/10.1017/S0269888900008122
The Knowledge Engineering Review
1,995
Michael Wooldridge; Nicholas R. Jennings
Surveys the properties, architectures, and engineering approaches that distinguish autonomous agents from ordinary software modules.
Grounds the agency-at-the-nodes boundary that separates an agent graph from a deterministic workflow.
Peer-reviewed research
Roles
age-0005
Research Foundations
Shared-state architectures
Paper
The Blackboard Model of Problem Solving and the Evolution of Blackboard Architectures
https://doi.org/10.1609/aimag.v7i2.537
AI Magazine
1,986
H. Penny Nii
Describes systems in which independent specialists coordinate opportunistically through a shared problem state and a control component.
Supplies a durable model for shared state without requiring every node to exchange its full context directly.
Peer-reviewed research
State
age-0006
Research Foundations
Learned communication
Paper
Learning to Communicate with Deep Multi-Agent Reinforcement Learning
https://proceedings.neurips.cc/paper_files/paper/2016/hash/c7635bfd99248a2cdef8249ef7bfbef4-Abstract.html
NeurIPS
2,016
Jakob Foerster; Ioannis Alexandros Assael; Nando de Freitas; Shimon Whiteson
Introduces reinforcement-learning methods that let agents learn communication protocols alongside their task policies, including discrete messages for execution.
Shows that edge content and communication policy can be engineered or learned rather than treated as free-form chat.
Peer-reviewed research
Handoffs
age-0007
Research Foundations
Learned communication
Paper
TarMAC: Targeted Multi-Agent Communication
https://proceedings.mlr.press/v97/das19a.html
ICML
2,019
Abhishek Das et al.
Uses attention to let agents address different messages to selected recipients instead of broadcasting the same information to the whole team.
Motivates selective, recipient-aware handoffs when all-to-all communication is wasteful or distracting.
Peer-reviewed research
Handoffs
age-0008
Research Foundations
LLM multi-agent systems
Paper
AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversations
https://openreview.net/forum?id=BAakY1hNKS
COLM
2,024
Qingyun Wu; Gagan Bansal; Jieyu Zhang; Yiran Wu; Beibin Li; Erkang Zhu; Li Jiang; Xiaoyun Zhang; Shaokun Zhang; Jiale Liu; Ahmed Hassan Awadallah; Ryen W. White; Doug Burger; Chi Wang
Presents a framework for composing customizable conversational agents that can combine language models, tools, code execution, and human input.
An early, influential demonstration that agent roles and conversation links can be expressed as an executable topology.
Peer-reviewed research
Topology
age-0009
Research Foundations
Topology optimization
Paper
GPTSwarm: Language Agents as Optimizable Graphs
https://proceedings.mlr.press/v235/zhuge24a.html
ICML
2,024
Mingchen Zhuge et al.
Represents language-agent systems as computational graphs and optimizes graph components from task feedback.
Makes the graph itself an optimization target rather than a fixed orchestration diagram.
Peer-reviewed research
Evolution
age-0010
Research Foundations
Dynamic topology
Paper
A Dynamic LLM-Powered Agent Network for Task-Oriented Agent Collaboration
https://openreview.net/forum?id=XII0Wp1XA9
COLM
2,024
Zijun Liu et al.
Constructs task-specific collaboration networks that can vary which agents participate and how they communicate instead of relying on one fixed team.
Provides evidence for adapting the work graph to the task while keeping the available agent roles reusable.
Peer-reviewed research
Evolution
age-0011
Research Foundations
Communication efficiency
Paper
Cut the Crap: An Economical Communication Pipeline for LLM-based Multi-Agent Systems
https://openreview.net/forum?id=LkzuPorQ5L
ICLR
2,025
Guibin Zhang et al.
Studies an economical communication pipeline that reduces redundant information exchanged among language-model agents.
Treats edge traffic as a measurable cost and tests whether less communication can preserve useful collaboration.
Peer-reviewed research
Observability & cost
age-0012
Research Foundations
Topology optimization
Paper
G-Designer: Architecting Multi-agent Communication Topologies via Graph Neural Networks
https://proceedings.mlr.press/v267/zhang25cu.html
ICML
2,025
Guibin Zhang et al.
Uses graph neural networks to design communication structures for multi-agent systems instead of assuming a complete or manually chosen graph.
Connects task performance to explicit topology search and exposes communication structure as an engineering variable.
Peer-reviewed research
Evolution
age-0013
Research Foundations
System search
Paper
Automated Design of Agentic Systems
https://openreview.net/forum?id=t9U3LW7JVX
ICLR
2,025
Shengran Hu; Cong Lu; Jeff Clune
Uses a meta-agent to propose, evaluate, and iteratively improve code-defined agentic systems across tasks.
Demonstrates automated search over coordination logic while retaining executable artifacts that engineers can inspect.
Peer-reviewed research
Evolution
age-0014
Research Foundations
Workflow search
Paper
AFlow: Automating Agentic Workflow Generation
https://openreview.net/forum?id=z5uVAKwmjf
ICLR
2,025
Jiayi Zhang et al.
Searches over reusable workflow operators to generate task-specific agentic workflows and improve them from evaluation results.
Offers a concrete method for evolving work graphs against measurable objectives rather than intuition alone.
Peer-reviewed research
Evolution
age-0015
Research Foundations
Joint optimization
Paper
Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies
https://openreview.net/forum?id=I05H9RUzHB
ICLR
2,026
Han Zhou et al.
Studies joint optimization of agent prompts and communication topology instead of tuning either component in isolation.
Shows that node behavior and graph structure interact and may need to evolve together.
Peer-reviewed research
Evolution
age-0016
Research Foundations
Debate and councils
Paper
Improving Factuality and Reasoning in Language Models through Multiagent Debate
https://proceedings.mlr.press/v235/du24e.html
ICML
2,024
Yilun Du et al.
Tests rounds of proposal and critique among multiple language-model instances as a way to improve factual and reasoning answers.
Supplies an empirical basis for debate-style gates while leaving room to examine correlated errors and added cost.
Peer-reviewed research
Gates
age-0017
Research Foundations
Debate and councils
Paper
Improving Multi-Agent Debate with Sparse Communication Topology
https://aclanthology.org/2024.findings-emnlp.427/
Findings of EMNLP
2,024
Yunxuan Li et al.
Examines multi-agent debate under sparse communication structures rather than defaulting to full information exchange among every participant.
Isolates topology as a factor in debate quality and communication efficiency.
Peer-reviewed research
Topology
age-0018
Research Foundations
Feedback and memory
Paper
Reflexion: Language Agents with Verbal Reinforcement Learning
https://openreview.net/forum?id=vAElhFcKW6
NeurIPS
2,023
Noah Shinn et al.
Lets an agent convert feedback into textual reflections stored in episodic memory and reused on later attempts.
Clarifies how a node loop can persist learning across retries without changing model weights.
Peer-reviewed research
State
age-0019
Research Foundations
Tool-grounded verification
Paper
CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing
https://openreview.net/forum?id=Sx038qxjek
ICLR
2,024
Zhibin Gou et al.
Uses external tools to obtain feedback that a language model can apply when critiquing and revising its outputs.
Supports verification gates grounded in observable evidence instead of another ungrounded model opinion.
Peer-reviewed research
Gates
age-0020
Production Case Studies
Generalist agent team
Paper
Magentic-One: A Generalist Multi-Agent System for Solving Complex Tasks
https://www.microsoft.com/en-us/research/publication/magentic-one-a-generalist-multi-agent-system-for-solving-complex-tasks/
Microsoft Research MSR-TR-2024-47
2,024
Adam Fourney et al.
Documents an orchestrator-led team of specialized agents for web, file, coding, and terminal tasks, with progress tracking and replanning.
Provides a concrete generalist topology whose role boundaries and recovery behavior can be examined as a system.
Research preprint
Roles
age-0021
Production Case Studies
Parallel research
Blog
How we built our multi-agent research system
https://www.anthropic.com/engineering/multi-agent-research-system
Anthropic Engineering
2,025
Jeremy Hadfield et al.
Describes a lead research agent that creates parallel subagents, delegates searches, and synthesizes their findings, including operational lessons from production.
A detailed case study of dynamic fan-out and fan-in, context separation, evaluation, and the token cost of an adaptive work graph.
Practitioner analysis
Work graphs
age-0022
Frameworks & SDKs
Role-based teams
Docs
CrewAI Crews
https://docs.crewai.com/en/concepts/crews
CrewAI Documentation
2,026
CrewAI
Documents role-based agent crews, assigned tasks, delegation, and sequential or hierarchical execution processes.
Offers accessible primitives for testing explicit ownership and manager-worker coordination.
Official documentation
Roles
age-0023
Frameworks & SDKs
Agent orchestration
Docs
OpenAI Agents SDK: Agent orchestration
https://openai.github.io/openai-agents-python/multi_agent/
OpenAI Agents SDK Documentation
2,026
OpenAI
Explains manager-style orchestration with agents exposed as tools and decentralized orchestration through handoffs.
Makes the centralized-versus-decentralized topology choice explicit in a production SDK.
Official documentation
Topology
age-0024
Frameworks & SDKs
Pattern catalog
Docs
Strands Agents: Multi-Agent Patterns
https://strandsagents.com/docs/user-guide/concepts/multi-agent/multi-agent-patterns/
Strands Agents Documentation
2,026
Strands Agents; AWS
Documents several coordination shapes for composing agents, including supervisor, swarm, workflow, and graph-oriented patterns.
Lets builders compare topology choices within one SDK instead of treating one pattern as universal.
Official documentation
Topology
age-0025
Frameworks & SDKs
Typed agents
Docs
Pydantic AI: Multi-Agent Applications
https://pydantic.dev/docs/ai/guides/multi-agent-applications/
Pydantic AI Documentation
2,026
Pydantic
Shows delegation, programmatic control flow, and graph-based state machines for composing typed Python agents.
Useful for expressing node inputs, outputs, dependencies, and orchestration boundaries in ordinary application code.
Official documentation
Topology
age-0026
Frameworks & SDKs
Multi-agent orchestration
Docs
LlamaIndex: Multi-Agent Patterns
https://developers.llamaindex.ai/python/framework/understanding/agent/multi_agent/
LlamaIndex Documentation
2,026
LlamaIndex
Covers agent workflow, orchestrator, and planner-oriented approaches for coordinating specialized agents.
Provides implementation patterns for choosing who owns delegation and how results return to the coordinating node.
Official documentation
Handoffs
age-0027
Frameworks & SDKs
Graph workflows
Docs
Google ADK: Graph-based Agent Workflows
https://adk.dev/graphs/
Google ADK Documentation
2,026
Google
Documents declarative workflows whose nodes combine agents, tools, functions, and human input through explicit edges, typed data passing, routing, branching, state, fan-out and join, loops, escalation, and nesting.
Makes the graph load-bearing and inspectable while separating deterministic process control from model reasoning.
Official documentation
Work graphs
age-0028
Frameworks & SDKs
Graph workflows
Docs
Microsoft Agent Framework: Workflows
https://learn.microsoft.com/en-us/agent-framework/workflows/
Microsoft Learn
2,026
Microsoft
Describes workflows built from executors and explicit edges, with support for branching, aggregation, state, and checkpointing.
Exposes the work graph as an inspectable program rather than hiding coordination inside prompts.
Official documentation
Work graphs
age-0029
Frameworks & SDKs
Graph runtime
Docs
LangGraph overview
https://docs.langchain.com/oss/python/langgraph/overview
LangGraph Documentation
2,025
LangChain
Introduces a low-level runtime for stateful agent graphs with durable execution, streaming, memory, and human intervention.
A widely used substrate for implementing explicit nodes, edges, state transitions, and resumable work graphs.
Official documentation
Work graphs
age-0030
Protocols & Handoffs
In-process transfer
Docs
OpenAI Agents SDK: Handoffs
https://openai.github.io/openai-agents-python/handoffs/
OpenAI Agents SDK Documentation
2,026
OpenAI
Documents transfers from one agent to another, including tool-shaped handoff schemas, input filters, and callbacks.
Turns an edge into an explicit contract controlling when ownership moves and what context crosses with it.
Official documentation
Handoffs
age-0031
Protocols & Handoffs
Tool and context protocol
Standard
Model Context Protocol Specification 2025-11-25
https://modelcontextprotocol.io/specification/2025-11-25
MCP Specification 2025-11-25
2,025
Model Context Protocol project; Agentic AI Foundation
Specifies a client-server protocol through which AI applications discover and use tools, resources, prompts, and contextual data.
Standardizes capability and context edges, while remaining distinct from a protocol for delegating work between autonomous agents.
Industry standard
Handoffs
age-0032
Protocols & Handoffs
Agent interoperability
Standard
Agent2Agent Protocol Specification v1.0.0
https://a2a-protocol.org/v1.0.0/specification/
A2A Specification
2,026
A2A Protocol Working Group; Linux Foundation
Defines interoperable agent discovery, task lifecycle, messages, artifacts, streaming, asynchronous updates, version negotiation, and multiple protocol bindings across service boundaries.
Provides a version-pinned wire contract for cross-system handoffs where agents cannot share an in-process runtime.
Industry standard
Handoffs
age-0033
State, Memory & Artifacts
Checkpointing
Docs
LangGraph Persistence
https://docs.langchain.com/oss/python/langgraph/persistence
LangGraph Documentation
2,025
LangChain
Documents thread-scoped checkpoints, saved state, replay, state inspection, and memory storage for LangGraph runs.
Shows how graph state can become the recoverable system of record instead of living only in model context.
Official documentation
State
age-0034
State, Memory & Artifacts
Conversation state
Docs
OpenAI Agents SDK: Sessions
https://openai.github.io/openai-agents-python/sessions/
OpenAI Agents SDK Documentation
2,026
OpenAI
Documents persistent conversation history that can be loaded and updated across repeated agent runs.
Provides a bounded mechanism for carrying state across nodes and turns without manually rebuilding every prompt.
Official documentation
State
age-0035
Verification & Evals
Workflow evaluation
Docs
Evaluate agent workflows
https://developers.openai.com/api/docs/guides/agent-evals
OpenAI API Documentation
2,026
OpenAI
Explains how to build datasets, graders, trace-based evaluations, and reproducible checks for agent workflows.
Makes gates testable at both the final outcome and the intermediate handoff level.
Official documentation
Gates
age-0036
Verification & Evals
Evaluation practice
Blog
Demystifying evals for AI agents
https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents
Anthropic Engineering
2,026
Mikaela Grace et al.
Presents practical guidance for defining tasks, outcomes, graders, and evaluation suites for agents with non-deterministic trajectories.
Helps turn vague reviewer judgments into evidence gates that can guide graph changes.
Practitioner analysis
Gates
age-0037
Verification & Evals
Evaluation framework
Tool
Inspect AI
https://inspect.aisi.org.uk/
Inspect AI Documentation
2,024
UK AI Security Institute
Provides an open-source evaluation framework with tasks, solvers, scorers, sandboxed tools, and structured logs.
Supports reproducible gate nodes and traceable evidence for both individual agents and composed systems.
Maintained OSS project
Gates
age-0038
Reliability & Durable Execution
Durable actors and workflows
Docs
Dapr Agents introduction
https://docs.dapr.io/developing-ai/dapr-agents/dapr-agents-introduction/
Dapr Documentation
2,026
Dapr; CNCF
Introduces an agent framework built on durable actors and workflows, with state, messaging, and recovery supplied by the Dapr runtime.
Shows how graph nodes can inherit distributed-systems durability instead of implementing recovery only in prompts.
Official documentation
Reliability
age-0039
Reliability & Durable Execution
Durable execution integration
Tool
Temporal integration for OpenAI Agents SDK
https://github.com/temporalio/sdk-python/tree/main/temporalio/contrib/openai_agents
Temporal Python SDK
2,025
Temporal
Integrates OpenAI agent runs, model calls, and tools with Temporal workflows and activities for durable execution.
Provides replay, retry, timeout, and recovery semantics beneath an agent graph without asking the model to manage them.
Maintained OSS project
Reliability
age-0040
Reliability & Durable Execution
Database-backed workflows
Docs
DBOS AI Quickstart
https://docs.dbos.dev/ai/ai-quickstart
DBOS Documentation
2,026
DBOS
Shows how to place AI application steps inside durable workflows whose progress is recorded and recoverable after interruption.
Offers a compact path from an agent prototype to resumable execution with explicit step boundaries.
Official documentation
Reliability
age-0041
Reliability & Durable Execution
Durable agent patterns
Docs
Restate Durable Agents
https://docs.restate.dev/ai/patterns/durable-agents
Restate Documentation
2,026
Restate
Documents durable agent patterns using persisted execution, reliable calls, retries, timers, and stateful services.
Maps common mid-graph failures to runtime guarantees rather than fragile application-level retry code.
Official documentation
Reliability
age-0042
Observability & Cost
Tracing
Docs
OpenAI Agents SDK: Tracing
https://openai.github.io/openai-agents-python/tracing/
OpenAI Agents SDK Documentation
2,026
OpenAI
Documents traces and spans for agent runs, model generations, tool calls, handoffs, and guardrails.
Makes node and edge behavior inspectable so latency, failures, and expensive paths can be attributed correctly.
Official documentation
Observability & cost
age-0043
Observability & Cost
Telemetry standard
Standard
OpenTelemetry GenAI Semantic Conventions
https://github.com/open-telemetry/semantic-conventions-genai
OpenTelemetry GenAI repository
2,026
OpenTelemetry GenAI SIG
Develops shared telemetry names and attributes for generative-AI model, tool, and agent operations in traces, metrics, and events.
Helps graph telemetry remain portable across runtimes and observability vendors.
Maintained OSS project
Observability & cost
age-0044
Observability & Cost
Cost attribution
Docs
Arize Phoenix: Cost Tracking
https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking
Arize Phoenix Documentation
2,026
Arize AI
Explains how Phoenix derives and displays token usage and model cost from traced generative-AI calls.
Supports per-node and per-path cost accounting instead of treating a multi-agent run as one opaque bill.
Official documentation
Observability & cost
age-0045
Observability & Cost
Observability platform
Docs
LangSmith Observability
https://docs.langchain.com/langsmith/observability
LangSmith Documentation
2,026
LangChain
Documents tracing, dashboards, alerts, feedback, and evaluation views for language-model and agent applications.
Provides operational views for following execution across nodes and locating failures on the critical path.
Official documentation
Observability & cost
age-0046
Benchmarks & Datasets
Collaboration benchmark
Benchmark
MultiAgentBench: Evaluating Collaboration and Competition of LLM Agents
https://aclanthology.org/2025.acl-long.421/
ACL
2,025
Kunlun Zhu et al.
Benchmarks language-model agents in collaborative and competitive settings while examining coordination processes as well as task outcomes.
Measures properties of the team interaction that single-agent benchmarks cannot expose.
Benchmark/dataset
Observability & cost
age-0047
Benchmarks & Datasets
Failure diagnosis
Benchmark
Why Do Multi-Agent LLM Systems Fail?
https://nips.cc/virtual/2025/poster/121528
NeurIPS Datasets & Benchmarks
2,025
Mert Cemri et al.
Provides a structured taxonomy and evaluation approach for diagnosing coordination failures in multi-agent language-model systems.
Turns reliability incidents into recurring, attributable failure classes that can guide graph redesign.
Benchmark/dataset
Reliability
age-0048
Critiques & Limits
Self-correction limits
Paper
Large Language Models Cannot Self-Correct Reasoning Yet
https://openreview.net/forum?id=IkmD3fKBPQ
ICLR
2,024
Jie Huang et al.
Finds that intrinsic self-correction without reliable external feedback often fails to improve reasoning and can reduce accuracy.
Warns against using an ungrounded critic node as evidence simply because it is separate from the producing node.
Peer-reviewed research
Gates
age-0049
Critiques & Limits
Scaling evidence
Paper
Towards a Science of Scaling Agent Systems
https://arxiv.org/abs/2512.08296
arXiv; Google Research
2,025
Yubin Kim et al.
Studies how agent-system performance changes across tasks, models, coordination structures, and scaling choices under controlled experiments.
Tests the assumption that adding agents reliably helps and frames scaling as an empirical topology decision.
Research preprint
Topology
age-0050
Critiques & Limits
Token-budget comparison
Paper
Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets
https://arxiv.org/abs/2604.02460
arXiv
2,026
Dat Tran; Douwe Kiela
Compares single-agent and multi-agent approaches to multi-hop reasoning while holding the total thinking-token budget constant.
Provides the cost-controlled baseline needed before claiming that coordination, rather than extra inference, caused an improvement.
Research preprint
Topology
age-0051
Start Here
Contemporary framing
Blog
From Loop Engineering to Graph Engineering?
https://x.com/IntuitMachine/status/2078419526354378975
X Articles
2,026
Carlos E. Perez
Frames the shift as loop architecture: networks of improvement cycles that monitor, feed, constrain, and correct one another, with reliability located in their edges.
Adds the grounding requirement missing from topology-only accounts: independent counter-metrics, frozen tests or rules, external anchors, and human ownership of root objectives.
Practitioner analysis
Gates
age-0052
Research Foundations
Blackboard coordination
Paper
A Multi-Level Organization for Problem Solving Using Many, Diverse, Cooperating Sources of Knowledge
https://www.ijcai.org/Proceedings/75/Papers/072.pdf
IJCAI
1,975
Lee D. Erman; Victor R. Lesser
Presents the Hearsay-II multi-level blackboard, where independent knowledge sources react to shared hypotheses, create explicit structural dependencies, and verify or revise one another's contributions.
Provides an early architecture for loosely coupled specialist nodes coordinating through inspectable shared state instead of direct all-to-all calls.
Peer-reviewed research
State
age-0053
Research Foundations
Negotiated delegation
Paper
The Contract Net Protocol: High-Level Communication and Control in a Distributed Problem Solver
https://doi.org/10.1109/TC.1980.1675516
IEEE Transactions on Computers
1,980
Reid G. Smith
Defines a negotiation protocol in which managers announce tasks, potential contractors bid, and awards establish temporary problem-solving relationships.
Supplies the classic contract for capability-aware delegation and auditable assignment edges between autonomous nodes.
Peer-reviewed research
Handoffs
age-0054
Start Here
Communication survey
Paper
The Five Ws of Multi-Agent Communication: Who Talks to Whom, When, What, and Why – A Survey from MARL to Emergent Language and LLMs
https://openreview.net/pdf?id=LGsed0QQVq
Transactions on Machine Learning Research
2,026
Jingdi Chen; Hanqing Yang; Zongjun Liu; Carlee Joe-Wong
Synthesizes multi-agent communication across reinforcement learning, emergent language, and LLM systems through sender, recipient, timing, content, and purpose decisions.
Provides a design-oriented map for engineering edge selection, message timing, payloads, grounding, scalability, and interpretability.
Peer-reviewed research
Handoffs
age-0055
Research Foundations
Collaboration scaling
Paper
Scaling Large Language Model-based Multi-Agent Collaboration
https://openreview.net/forum?id=K3n5jPkrU6
ICLR
2,025
Chen Qian; Zihao Xie; YiFei Wang; Wei Liu; Kunlun Zhu; Hanchen Xia; Yufan Dang; Zhuoyun Du; Weize Chen; Cheng Yang; Zhiyuan Liu; Maosong Sun
Introduces MacNet, a DAG-based collaboration architecture executed in topological order, and studies communication structure while scaling experiments beyond 1,000 agents.
Makes topology a causal scaling variable rather than assuming that larger teams or denser communication automatically help.
Peer-reviewed research
Topology
age-0056
Research Foundations
Holistic orchestration
Paper
MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks
https://openreview.net/forum?id=3fGXBm4c3S
ICML
2,026
Zixuan Ke et al.
Formulates orchestration as reinforcement-learned generation of a complete multi-agent program and evaluates it across controlled dimensions including depth, horizon, breadth, parallelism, and robustness.
Tests when whole-system graph structure helps instead of attributing gains to coordination without controlled task evidence.
Peer-reviewed research
Work graphs
age-0057
Research Foundations
Conditional topology
Paper
CARD: Towards Conditional Design of Multi-agent Topological Structures
https://openreview.net/forum?id=JgvJdICc6P
ICLR
2,026
Tongtong Wu et al.
Generates communication graphs conditioned on agent roles, models, tools, and data sources, and adapts topology as task resources change.
Treats the available capabilities and directed edges as a versionable organizational artifact rather than a fixed team template.
Peer-reviewed research
Evolution
age-0058
Reliability & Durable Execution
Resilient topology
Paper
ResMAS: Resilience Optimization in LLM-based Multi-agent Systems
https://ojs.aaai.org/index.php/AAAI/article/view/40824
AAAI
2,026
Zhilun Zhou et al.
Learns task-specific resilient communication topologies and topology-aware prompts after measuring how graph structure and node instructions affect performance under agent failures and other perturbations.
Moves resilience from reactive recovery into the design of the graph itself and evaluates transfer to new tasks and models.
Peer-reviewed research
Reliability
age-0059
Research Foundations
System evolution
Paper
EvoMAS: Evolutionary Generation of Multi-Agent Systems
https://openreview.net/forum?id=ic0AGRIkmY
ICML
2,026
Yuntong Hu et al.
Evolves structured multi-agent configurations through trace-guided mutation, crossover, selection, and an experience memory across reasoning, coding, and tool-use tasks.
Shows how an inspectable team specification can evolve from execution evidence while retaining executability and runtime robustness.
Peer-reviewed research
Evolution
age-0060
Critiques & Limits
Error propagation
Paper
Understanding the Information Propagation Effects of Communication Topologies in LLM-based Multi-Agent Systems
https://aclanthology.org/2025.emnlp-main.623/
EMNLP
2,025
Xu Shen et al.
Causally studies correct and erroneous information propagation across communication densities and finds that moderately sparse structures can preserve useful diffusion while suppressing errors.
Provides evidence against defaulting to dense graphs and links topology decisions to measured error amplification.
Peer-reviewed research
Topology
age-0061
Frameworks & SDKs
Graph-centric orchestration
Paper
MASFactory: A Graph-centric Framework for Orchestrating LLM-Based Multi-Agent Systems with Vibe Graphing
https://aclanthology.org/2026.acl-demo.35/
ACL System Demonstrations
2,026
Yang Liu et al.
Compiles natural-language intent into an editable workflow specification and executable directed graph, with reusable components, topology preview, runtime tracing, multimodal messages, and human interaction.
Provides a direct implementation path from an inspectable organizational graph to execution and evaluates it on seven public benchmarks.
Peer-reviewed research
Work graphs
age-0062
Frameworks & SDKs
Directed agent graphs
Docs
AutoGen GraphFlow (Workflows)
https://microsoft.github.io/autogen/dev/user-guide/agentchat-user-guide/graph-flow.html
Microsoft AutoGen Documentation
2,026
Microsoft
Implements directed multi-agent execution graphs with sequential, parallel, conditional, fan-in, and cyclic paths, edge conditions, activation groups, safe loop exits, and separately configurable message filtering. The feature is explicitly experimental.
Distinguishes the execution graph from the message graph, exposing both who acts next and what context each agent receives.
Official documentation
Work graphs
age-0063
Protocols & Handoffs
Durable remote tasks
Standard
Model Context Protocol: Tasks
https://modelcontextprotocol.io/specification/2025-11-25/basic/utilities/tasks
MCP Specification 2025-11-25
2,025
Model Context Protocol project; Agentic AI Foundation
Specifies experimental durable asynchronous request state machines with capability negotiation, polling, deferred results, progress, input-required states, cancellation, TTLs, and task-message correlation.
Turns a remote tool or context edge into a recoverable task contract that can outlive one synchronous request.
Official documentation
Reliability
age-0064
Protocols & Handoffs
Secure agent messaging
Docs
Secure Low-Latency Interactive Messaging (SLIM)
https://datatracker.ietf.org/doc/draft-mpsb-agntcy-slim/
IETF Datatracker; individual Internet-Draft
2,026
Luca Muscariello; Michele Papalini; Mauro Sardara; Sam Betts
Proposes a transport layer for A2A and MCP using gRPC over HTTP/2 and HTTP/3 with stream multiplexing, flow control, group communication, native RPC semantics, and MLS end-to-end encryption. It is an individual informational Internet-Draft with no formal IETF standing.
Adds a concrete secure transport substrate for high-volume graph edges that must cross process and organizational boundaries.
Official documentation
Handoffs
age-0065
Protocols & Handoffs
Agent identity and authorization
Docs
Accelerating the Adoption of Software and Artificial Intelligence Agent Identity and Authorization
https://csrc.nist.gov/pubs/other/2026/02/05/accelerating-the-adoption-of-software-and-ai-agent/ipd
NIST NCCoE Initial Public Draft
2,026
Harold Booth; William Fisher; Ryan Galluzzo; Joshua Roberts
Outlines considerations and open questions for standards-based identity, authorization, auditing, and non-repudiation when software and AI agents access enterprise systems and take actions. The concept paper remains an initial public draft under review.
Grounds graph roles and permissions in identity practice so delegation does not silently transfer more authority than an edge contract allows.
Official documentation
Roles
age-0066
State, Memory & Artifacts
Versioned work products
Docs
Google ADK: Artifacts
https://adk.dev/artifacts/
Google ADK Documentation
2,026
Google
Defines named, automatically versioned binary work products that agents and tools can save, load, list, and exchange within session-scoped or persistent user-scoped namespaces.
Provides explicit, inspectable edge artifacts instead of forcing large or structured outputs through conversational context.
Official documentation
State
age-0067
State, Memory & Artifacts
Procedural memory
Paper
LEGOMem: Modular Procedural Memory for Multi-agent LLM Systems for Workflow Automation
https://www.microsoft.com/en-us/research/publication/legomem-modular-procedural-memory-for-multi-agent-llm-systems-for-workflow-automation/
AAMAS
2,026
Dongge Han; Camille Couturier; Daniel Madrigal; Xuchao Zhang; Victor Ruehle; Saravan Rajmohan
Decomposes execution trajectories into reusable procedural memories and allocates them to an orchestrator and specialist agents for workflow automation.
Shows how durable experience can improve decomposition and delegation without collapsing every node's memory into one undifferentiated store.
Peer-reviewed research
State
age-0068
Verification & Evals
Adversarial agent detection
Paper
When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems
https://openreview.net/forum?id=BnduUW8izq
ICML
2,026
Haowen Xu et al.
Evaluates activation-space detection and restorative steering of compromised agents across five attack scenarios, multiple models, and synchronous and asynchronous multi-agent interactions.
Tests a topology-agnostic gate for locating and repairing a malicious node when its messages appear superficially benign.
Peer-reviewed research
Reliability
age-0069
Reliability & Durable Execution
Intervention-driven debugging
Paper
DoVer: Intervention-Driven Auto Debugging for LLM Multi-Agent Systems
https://iclr.cc/virtual/2026/poster/10007537
ICLR
2,026
Ming Ma et al.
Tests failure hypotheses by editing messages or plans and measuring whether each intervention repairs the outcome or advances execution in branching multi-agent traces.
Turns graph debugging into causal repair experiments instead of relying on plausible but unverified post-hoc explanations.
Peer-reviewed research
Reliability
age-0070
Reliability & Durable Execution
Durable graph workflows
Docs
Microsoft Agent Framework: Durable Extension
https://learn.microsoft.com/en-us/agent-framework/integrations/durable-extension
Microsoft Learn
2,026
Microsoft
Adds persisted sessions, checkpointed progress, failure recovery, external-event waits, and distributed hosting to agents, multi-agent orchestrations, and graph workflows. The documented packages remain prerelease.
Shows how an agent graph can resume without losing context or repeating completed work after interruption.
Official documentation
Reliability
age-0071
Critiques & Limits
Control-flow security
Paper
Breaking and Fixing Defenses Against Control Flow Hijacking in Multi-Agent Systems
https://openreview.net/forum?id=PNU9Rj5RDQ
ICLR
2,026
Rishi Dev Jha; Harold Triedman; Justin Wagle; Vitaly Shmatikov
Demonstrates attacks against alignment-check defenses and introduces ControlValve, which generates permitted control-flow graphs and enforces least privilege for each agent invocation.
Makes invocation authority and contextual permissions explicit on every edge instead of trusting a separate checker that can also be hijacked.
Peer-reviewed research
Reliability
age-0072
Observability & Cost
Budget-aware topology
Paper
BAMAS: Structuring Budget-Aware Multi-Agent Systems
https://ojs.aaai.org/index.php/AAAI/article/view/40226
AAAI
2,026
Liming Yang; Junyu Luo; Xuanzhe Liu; Yiling Lou; Zhenpeng Chen
Selects an LLM team with integer programming and then learns its collaboration topology under an explicit budget, reporting comparable performance with cost reductions of up to 86% on three tasks.
Jointly engineers node selection, edge structure, and spend instead of optimizing quality while treating inference cost as an afterthought.
Peer-reviewed research
Observability & cost
age-0073
Observability & Cost
Workflow reconstruction
Paper
AgentXRay: White-Boxing Agentic Systems via Workflow Reconstruction
https://arxiv.org/abs/2602.05353
ICML
2,026
Ruijie Shi; Houbin Zhang; Yuecheng Han; Yuheng Wang; Jingru Fan; Runde Yang; Yufan Dang; Huatao Li; Dewen Liu; Yuan Cheng; Chen Qian
Reconstructs an editable explicit stand-in workflow for a black-box agentic system from input-output behavior using iterative search and evaluation.
Offers a path to inspect, compare, and modify systems whose load-bearing workflow is hidden behind an API.
Peer-reviewed research
Observability & cost
age-0074
Benchmarks & Datasets
Distributed coordination
Benchmark
SILO-BENCH: A Scalable Environment for Evaluating Distributed Coordination in Multi-Agent LLM Systems
https://aclanthology.org/2026.acl-long.1354/
ACL
2,026
Yuzhe Zhang et al.
Benchmarks free-form coordination under information silos across 30 exact-answer tasks, three communication protocols, six agent scales, and three models while recording success, tokens, and communication density.
Tests whether more nodes and denser edges overcome distributed information constraints under explicit coordination and cost measures.
Benchmark/dataset
Topology
age-0075
Benchmarks & Datasets
Dynamic asynchronous evaluation
Benchmark
Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments
https://openreview.net/forum?id=9gw03JpKK4
ICLR
2,026
Romain Froger et al.
Evaluates agents in asynchronous simulated environments across execution, search, ambiguity, adaptation, temporal reasoning, noise, and agent-to-agent collaboration, with action-level verifiers and structured traces.
Makes delays, environmental change, peer communication, and externally checked writes first-class work-graph conditions.
Benchmark/dataset
Work graphs
age-0076
Benchmarks & Datasets
Adversarial multi-agent safety
Benchmark
TAMAS: Benchmarking Adversarial Risks in Multi-Agent LLM Systems
https://aclanthology.org/2026.acl-long.1442/
ACL
2,026
Ishan Kavathekar; Hemang Jain; Ameya Rathod; Ponnurangam Kumaraguru; Tanuja Ganu
Provides five scenarios with 300 adversarial instances, six attack types, 211 tools, 100 harmless tasks, and multiple AutoGen and CrewAI interaction configurations, plus an Effective Robustness Score.
Evaluates whether safety controls preserve useful work while attacks exploit the distinctive trust and communication surfaces of a multi-agent graph.
Benchmark/dataset
Reliability
age-0077
Benchmarks & Datasets
Failure attribution
Benchmark
Seeing the Whole Elephant: A Benchmark for Failure Attribution in LLM-based Multi-Agent Systems
https://aclanthology.org/2026.acl-long.912/
ACL
2,026
Mengzhuo Chen et al.
Introduces TraceElephant with full execution traces and reproducible environments for attributing failures to responsible agents and decisive steps.
Measures causal diagnosis over nodes, messages, plans, and dependencies instead of evaluating only the final team output.
Benchmark/dataset
Observability & cost
age-0078
Production Case Studies
Secure delegated access
Blog
Creating AI agent solutions for warehouse data access and security
https://engineering.fb.com/2025/08/13/data-infrastructure/agentic-solution-for-warehouse-data-access/
Engineering at Meta
2,025
Can Lin; Uday Ramesh Savagaonkar; Iuliu Rus; Komal Mangtani
Describes collaborating data-user and data-owner agents, specialized subagents, triage, permission negotiation, human oversight, access budgets, analytical risk rules, output guardrails, traces, and daily regression evaluation.
Shows plural bounded agency governed by independent rule-based gates rather than relying on model judgment for security decisions.
Practitioner analysis
Gates
age-0079
Production Case Studies
Staged agent swarm
Blog
How Meta Used AI to Map Tribal Knowledge in Large-Scale Data Pipelines
https://engineering.fb.com/2026/04/06/developer-tools/how-meta-used-ai-to-map-tribal-knowledge-in-large-scale-data-pipelines/
Engineering at Meta
2,026
Krishna Ganeriwal; Plawan Rath; Ashwini Verma
Reports a staged swarm of more than 50 explorer, analyst, writer, critic, fixer, upgrader, tester, and gap-filling tasks, with repeated review rounds and recurring automated refresh runs.
Provides a concrete fan-out, critique, repair, integration, and recurrence graph with explicit artifacts and quality gates; efficiency figures are company-reported and preliminary.
Practitioner analysis
Work graphs
age-0080
Production Case Studies
Multi-hop identity and provenance
Blog
Solving the Identity Crisis for AI Agents
https://www.uber.com/by/en/blog/solving-the-agent-identity-crisis/
Uber Engineering
2,026
Matt Mathew; Prasad Borole; Meng Huang; Sergey Burykin; Gaurav Goel; Bayard Walsh
Describes an internal agent mesh with registered identities, SPIRE-backed workload attestation, short-lived audience-scoped tokens for every hop, actor-chain provenance, MCP gateway enforcement, and a standardized A2A client.
Treats identity, delegated authority, and provenance as mandatory edge state across a multi-agent graph; adoption and latency metrics are company-reported.
Practitioner analysis
Handoffs
age-0081
Research Foundations
Task-adaptive topology synthesis
Paper
Dynamic Generation of Multi LLM Agents Communication Topologies with Graph Diffusion Models
https://aclanthology.org/2026.acl-long.1764/
ACL
2,026
Eric Hanchen Jiang et al.
Introduces Guided Topology Diffusion, which iteratively synthesizes sparse, task-adaptive communication graphs using a proxy model for objectives such as accuracy, utility, and cost.
Treats topology as a multi-objective, per-task design artifact rather than a static collaboration template.
Peer-reviewed research
Evolution
age-0082
Reliability & Durable Execution
Topology-conditioned memory leakage
Paper
Topology Matters: Measuring Memory Leakage in Multi-Agent LLMs
https://aclanthology.org/2026.findings-acl.1980/
Findings of ACL
2,026
Jinbo Liu et al.
Measures private-information leakage across six communication topologies, agent counts, and attacker-target placements, finding higher leakage with denser connectivity, shorter graph distance, and greater target centrality.
Makes privacy a measurable consequence of edge structure and node placement, motivating topology-aware access controls.
Peer-reviewed research
Reliability
age-0083
Critiques & Limits
Collaboration-induced diversity collapse
Paper
Diversity Collapse in Multi-Agent LLM Systems: Structural Coupling and Collective Failure in Open-Ended Idea Generation
https://aclanthology.org/2026.findings-acl.13/
Findings of ACL
2,026
Nuo Chen et al.
Studies multi-agent ideation across model, cognition, and system levels, finding diminishing group-size returns, authority effects, and faster premature convergence under dense communication.
Shows that interaction can contract a search space, so graph designs for exploration must preserve independence and meaningful disagreement.
Peer-reviewed research
Topology
age-0084
Research Foundations
Dynamic node and edge elimination
Paper
AgentDropout: Dynamic Agent Elimination for Token-Efficient and High-Performance LLM-Based Multi-Agent Collaboration
https://aclanthology.org/2025.acl-long.1170/
ACL
2,025
Zhexuan Wang; Yutong Wang; Xuebo Liu; Liang Ding; Miao Zhang; Jie Liu; Min Zhang
Optimizes communication-graph adjacency matrices across collaboration rounds to remove redundant agents and messages, reducing both prompt and completion token consumption in its evaluations.
Makes node participation and edge density explicit cost-quality decisions instead of assuming every role should run on every task.
Peer-reviewed research
Evolution
age-0085
Benchmarks & Datasets
Process-level collaboration
Benchmark
Collab-Overcooked: Benchmarking and Evaluating Large Language Models as Collaborative Agents
https://aclanthology.org/2025.emnlp-main.249/
EMNLP
2,025
Haochen Sun; Shuwen Zhang; Lujie Niu; Lei Ren; Hao Xu; Hao Fu; Fangkun Zhao; Caixia Yuan; Xiaojie Wang
Provides 30 open-ended cooperative tasks and process-oriented measures for goal interpretation, active collaboration, communication, and continuous adaptation across 13 language models.
Distinguishes reaching an answer from coordinating well, exposing collaboration failures that final-outcome metrics can hide.
Benchmark/dataset
Observability & cost
age-0086
Critiques & Limits
Inter-agent communication attacks
Paper
Red-Teaming LLM Multi-Agent Systems via Communication Attacks
https://aclanthology.org/2025.findings-acl.349/
Findings of ACL
2,025
Pengfei He; Yuping Lin; Shen Dong; Han Xu; Yue Xing; Hui Liu
Introduces an Agent-in-the-Middle attack that intercepts and manipulates inter-agent messages, evaluated across multiple frameworks, communication structures, and applications.
Shows that graph edges can compromise the whole system without altering its nodes, making message integrity and provenance part of the handoff contract.
Peer-reviewed research
Handoffs
age-0087
Research Foundations
Query-dependent architecture search
Paper
Multi-agent Architecture Search via Agentic Supernet
https://proceedings.mlr.press/v267/zhang25bi.html
ICML
2,025
Guibin Zhang; Luyang Niu; Junfeng Fang; Kun Wang; Lei Bai; Xiang Wang
Represents possible agentic architectures as a probabilistic supernet and samples query-dependent systems with tailored language-model calls, tool calls, and token costs across six benchmarks.
Replaces a one-size-fits-all graph with per-query architecture and resource allocation while keeping the design space explicit.
Peer-reviewed research
Evolution
age-0088
Critiques & Limits
Unequal agent contribution
Paper
Unlocking the Power of Multi-Agent LLM for Reasoning: From Lazy Agents to Deliberation
https://openreview.net/forum?id=5J6u03ObRZ
ICLR
2,026
Zhiwei Zhang et al.
Identifies lazy-agent behavior in which one role dominates a nominally collaborative system, then measures causal influence and uses verifiable rewards to encourage deliberation and selective reasoning restarts.
Requires builders to verify that each node contributes causal value rather than allowing a multi-agent graph to collapse into one effective agent.
Peer-reviewed research
Roles
age-0089
Benchmarks & Datasets
Multi-party negotiation
Benchmark
Cooperation, Competition, and Maliciousness: LLM-Stakeholders Interactive Negotiation
https://openreview.net/forum?id=59E19c6yrN
NeurIPS
2,024
Sahar Abdelnabi; Amr Gomaa; Sarath Sivaprasad; Lea Schönherr; Mario Fritz
Provides scorable, multi-agent, multi-issue negotiation games with metrics for task performance and role alignment, including cooperative, competitive, greedy, and adversarial participants.
Tests communication and collective decisions when graph nodes hold conflicting objectives or attempt manipulation rather than cooperating by default.
Benchmark/dataset
Handoffs
age-0090
Reliability & Durable Execution
Agentic application security risks
Standard
OWASP Top 10 for Agentic Applications for 2026
https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/
OWASP GenAI Security Project
2,025
OWASP GenAI Security Project; Agentic Security Initiative
Provides a globally peer-reviewed operational framework for critical risks in autonomous and agentic applications, developed with contributions from more than 100 security experts, researchers, and practitioners.
Supplies a practical threat checklist for node authority, tool permissions, memory provenance, inter-agent trust, and cascading failures across an agent graph.
Industry standard
Reliability
age-0091
Research Foundations
Task-adaptive graph routing
Paper
AMAS: Adaptively Determining Communication Topology for LLM-based Multi-agent System
https://aclanthology.org/2025.emnlp-industry.144/
EMNLP Industry Track
2,025
Hui Yi Leong; Yuheng Li; Yuqing Wu; Wenwen Ouyang; Wei Zhu; Jiechao Gao; Wei Han
Introduces a dynamic graph selector that uses lightweight LLM adaptation to choose task-specific communication structures and route each query through a matching agent pathway.
Operationalizes topology as an input-conditioned decision rather than a fixed template; evidence is currently concentrated in question answering, mathematics, and code generation.
Peer-reviewed research
Evolution
age-0092
Research Foundations
Collaboration-mode, role, and model routing
Paper
MasRouter: Learning to Route LLMs for Multi-Agent Systems
https://aclanthology.org/2025.acl-long.757/
ACL
2,025
Yanwei Yue; Guibin Zhang; Boyang Liu; Guancheng Wan; Kun Wang; Dawei Cheng; Yiyan Qi
Defines multi-agent system routing as a joint decision over collaboration mode, role allocation, and language-model selection, implemented with a cascaded controller that progressively constructs a task-specific system.
Makes node roles, model assignment, and whether collaboration is needed explicit routing decisions with measurable cost-quality trade-offs.
Peer-reviewed research
Evolution
age-0093
Research Foundations
Multi-hop evidence propagation
Paper
MOC: Multi-Order Communication in LLM-based Multi-Agent Systems
https://openreview.net/forum?id=wyynWicO5s
ICML
2,026
Yao Guan; Lin Wang; Zhihui Lu; Ziyi Wang; Wenzhu Yan; Qiang Duan
Constructs structured multi-order evidence streams so agents can receive relevant information from multiple upstream hops, then applies semantic-topological merging under token constraints.
Shows that graph engineering includes what evidence survives across paths, not only which nodes and edges exist; it improves communication over a supplied topology rather than selecting that topology.
Peer-reviewed research
Handoffs
age-0094
Research Foundations
Hierarchical node and topology optimization
Paper
HieraMAS: Optimizing Intra-Node LLM Mixtures and Inter-Node Topology for Multi-Agent Systems
https://openreview.net/forum?id=p7p5foAXaB
ICML
2,026
Tianjun Yao; Zhaoyi Li; Zhiqiang Shen
Models each functional role as a supernode containing heterogeneous language models in a propose-synthesis structure, then uses multi-level reward attribution and graph classification to select the inter-supernode topology.
Jointly engineers node composition, role capability, credit assignment, and communication edges instead of optimizing each dimension independently.
Peer-reviewed research
Evolution
age-0095
State, Memory & Artifacts
Hierarchical graph memory
Paper
GAM: Hierarchical Graph-based Agentic Memory for LLM Agents
https://aclanthology.org/2026.acl-long.1600/
ACL
2,026
Zhaofen Wu; Hanrong Zhang; Fulin Lin; Wujiang Xu; Xinran Xu; Yankai Chen; Henry Peng Zou; Shaowen Chen; Weizhi Zhang; Xue Liu; Philip S. Yu; Hongwei Wang
Separates rapid memory encoding from stable consolidation through an event-progression graph, a topic-associative network, and graph-guided multi-factor retrieval.
Provides a concrete graph contract for encoding, consolidating, connecting, and retrieving state; it does not address shared-memory permissions or concurrent writes among agents.
Peer-reviewed research
State
age-0096
State, Memory & Artifacts
Selective shared memory
Paper
Learning to Share: Selective Memory for Efficient Parallel Agentic Systems
https://openreview.net/forum?id=cCFyY2LmF5
ICML
2,026
Joseph Fioresi; Parth Parag Kulkarni; Ashmal Vayani; Song Wang; Mubarak Shah
Adds a global memory bank to parallel agent teams and trains an admission controller with stepwise reinforcement learning and usage-aware credit assignment to retain reusable intermediate work.
Treats cross-team memory writes as governed graph operations that can reduce duplicate work while limiting indiscriminate context growth.
Peer-reviewed research
State
age-0097
Observability & Cost
Graph-aware cache reuse
Paper
Accelerating Language Model Workflows with Prompt Choreography
https://aclanthology.org/2026.tacl-1.13/
TACL
2,026
TJ Bai; Jason Eisner
Executes multi-agent workflows with a dynamic global key-value cache in which each call can attend to a reordered subset of previously encoded messages, including parallel branches.
Makes workflow dependencies and reusable message state explicit execution concerns; its primary result concerns speed, and cache reuse can alter model behavior.
Peer-reviewed research
Work graphs
age-0098
Benchmarks & Datasets
Enterprise workflow benchmark
Benchmark
Can LLMs Help You at Work? A Sandbox for Evaluating LLM Agents in Enterprise Environments
https://aclanthology.org/2025.emnlp-main.466/
EMNLP
2,025
Harsh Vishwakarma; Ankush Agarwal; Ojas Patil; Chaitanya Devaguptapu; Mahesh Chandran
Introduces EnterpriseBench with 500 tasks across software engineering, HR, finance, and administration, including fragmented data, access-control hierarchies, and cross-functional workflows.
Tests work graphs that retrieve, modify, and transfer artifacts across services while respecting authority boundaries; the organization and services are simulated.
Benchmark/dataset
Work graphs
age-0099
Benchmarks & Datasets
Stateful tool-use evaluation
Benchmark
ToolSandbox: A Stateful, Conversational, Interactive Evaluation Benchmark for LLM Tool Use Capabilities
https://aclanthology.org/2025.findings-naacl.65/
Findings of NAACL
2,025
Jiarui Lu; Thomas Holleis; Yizhe Zhang; Bernhard Aumayer; Feng Nan; Haoping Bai; Shuang Ma; Shen Ma; Mengyu Li; Guoli Yin; Zirui Wang; Ruoming Pang
Evaluates stateful tool execution, implicit dependencies between tools, on-policy user interaction, and intermediate and final milestones over arbitrary trajectories.
Transfers milestone and state-transition testing to graph prerequisites, although the evaluated system is an agent-user-tool loop rather than an agent-agent topology.
Benchmark/dataset
State
age-0100
Benchmarks & Datasets
Trajectory-level evaluation
Benchmark
AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents
https://openreview.net/forum?id=4S8agvKjle
NeurIPS Datasets & Benchmarks
2,024
Chang Ma; Junlei Zhang; Zhihao Zhu; Cheng Yang; Yujiu Yang; Yaohui Jin; Zhenzhong Lan; Lingpeng Kong; Junxian He
Unifies partially observable, multi-round agent environments and adds a fine-grained progress-rate metric plus interactive trajectory analysis beyond final success.
Supplies milestone-level evaluation that can localize incomplete work, while its primarily single-agent tasks do not identify the responsible graph component by themselves.
Benchmark/dataset
Gates
End of preview. Expand in Data Studio

Awesome Graph Engineering Resource Atlas

A versioned collection of research, standards, frameworks, protocols, reliability systems, evaluations, and critiques for graph-structured multi-agent systems and programmable AI-agent organizations.

This dataset mirrors Awesome Graph Engineering. The GitHub JSONL file is canonical; the Hub exposes the same records through Dataset Viewer, direct downloads, datasets, and pandas.

Working definition

Graph engineering is the practice of specifying, executing, observing, and evolving a graph-structured agent system—its roles and runtime instances, the contracts that connect them, the state and artifacts they share, and the evidence by which their collective behavior is judged—so that the system can be controlled, tested, and improved as an engineered whole.

The graph must be load-bearing rather than decorative. Its declared topology, realized run graph, or graph-generating policy must materially constrain execution and remain inspectable enough to version, trace, evaluate, or deliberately change.

Graph engineering is used here as an emerging, non-standard term. The scope synthesizes multi-agent research, historical agent-system foundations, standards, runtime documentation, benchmarks, negative results, and practitioner analysis. The evidence map separates source-backed claims from analytical inferences. Carlos E. Perez’s “graph of loops” essay is one contemporary source, not an authority for the definition.

Dataset structure

The default configuration contains one train split. “Train” is the Hub's conventional single-split label; these rows are a resource atlas, not labeled examples or a model-training benchmark.

Each row has 13 fields:

Field Type Meaning
id string Stable age-NNNN resource identifier.
section string Reader-facing directory section.
subcategory string Specific discovery label within the section.
rtype string Resource format, such as Paper, Docs, Standard, or Benchmark.
title string Canonical resource title.
url string Canonical source URL.
venue string Publisher, venue, documentation site, or repository host.
year integer Publication year or year of the cited major release.
authors string Authors or maintaining organization.
description string Original summary of the resource's concrete contribution.
why string Practitioner rationale for including the resource.
evidence string Source-type label, not a quality score.
layer string Primary graph-engineering design layer.

resources.jsonl drives the Dataset Viewer. resources.csv is an equivalent convenience export in the same row and field order. resource.schema.json is the machine-readable JSON Schema for one record.

Interoperable access points:

Load the data

from datasets import load_dataset

dataset = load_dataset("cy0307/awesome-graph-engineering")
resources = dataset["train"]
print(resources.num_rows, resources.column_names)
import pandas as pd

url = (
    "https://huggingface.co/datasets/cy0307/awesome-graph-engineering/"
    "resolve/main/resources.csv"
)
resources = pd.read_csv(url)
print(resources.groupby("layer").size().sort_values(ascending=False))

Intended uses

  • discover primary sources and implementation references by engineering layer;
  • seed literature reviews and architecture comparisons, followed by reading the linked originals;
  • analyze how the collection is distributed across source types, years, sections, and layers;
  • build educational tools, resource browsers, or retrieval indexes; and
  • propose corrections and additions through the canonical GitHub repository.

Limitations and responsible use

  • The collection is selective and versioned; it is not an exhaustive scrape or systematic review.
  • evidence records publication form. It does not score correctness, replication, maintenance, safety, or endorsement.
  • Official documentation is authoritative about intended product behavior, not independent proof of reliability.
  • Practitioner analysis may define or challenge a boundary, but recency does not give it priority over primary evidence.
  • The field and its vocabulary are changing quickly; verify current documentation, publication status, licenses, and security posture before adoption.
  • Linked resources are catalogued by URL only; verify each resource’s rights and license before reuse.

Read the full curation methodology, scope boundaries, and dataset contract before drawing aggregate conclusions.

License

The dataset metadata, schema, original summaries, and repository-created assets are dedicated under CC0 1.0 Universal. Linked resources are not included in that dedication and remain subject to their own rights and licenses. CC0 does not waive trademark or patent rights and provides the work without warranties. Citation is appreciated for scholarly traceability but is not required by CC0.

Versioning and synchronization

GitHub’s data/resources.jsonl is canonical. The CSV, README tables, interactive atlas, and Hub mirror are generated from it. Stable IDs are never recycled. When HF_TOKEN is configured, accepted changes to main trigger validation and mirror publication.

Citation

Curated by He Chaoyue.

@misc{he2026awesomegraphengineering,
  author       = {He, Chaoyue},
  title        = {Awesome Graph Engineering: A Field Guide, Dataset, and Interactive Atlas for Programmable AI-Agent Organizations},
  year         = {2026},
  version      = {1.3.0},
  publisher    = {GitHub},
  howpublished = {\url{https://github.com/ChaoYue0307/awesome-graph-engineering/releases/tag/v1.3.0}},
  url          = {https://github.com/ChaoYue0307/awesome-graph-engineering/releases/tag/v1.3.0}
}

Machine-readable metadata is available in CITATION.cff. Cite the original linked works for claims derived from them.

Downloads last month
13