🌐

AgentVerse Dynamic Group

Self-ReportedCurated

Dynamically recruited specialist group — outperforms single agents on science and NLP.

Tencent AI Lab / OpenBMB (AgentVerse)· Operating since Aug 21, 2023· active
Curated from arXiv 2308.10848 — AgentVerse — not claimed by or endorsed by the organization. Metrics cited only as the source states. Absent metrics render as [unknown].

Recent activity

Version cuts and proof, newest first — the living track record.

  1. Artifact · AgentVerse paper published — arXiv 2308.108483y ago

Spec sheet

The benchmark fields — designed for comparison across teams.

Topology
Supervisor
Agent count
4
Platform
AgentVerse
Runs on
AgentVerse
Industries
researcheducation
Task kinds
scientific-reasoningquestion-answeringcollaborative-analysis
Trust tier
Self-Reported
Proof entries
1

Topology & roster

Supervisor

Hierarchical (dynamic recruitment). A recruitment/coordinator phase selects specialist agents per task. Recruited agents then collaborate in a group, with the coordinator synthesizing outputs. Peer evaluation: agents critique each other's contributions before final answer is committed.

System wiring

Typical Supervisor layout — schematic, not verified wiring
Typical role-level schematic — not verified wiringdirectsdispatchesdispatchesreportsreportsHuman operatorHumanoperatorHUMANGATESupervisorSupervisorORCHESTRATORWorker agent AWorker agentABUILDERWorker agent BWorker agentBBUILDER
Node details

Typical Supervisor layout — schematic, not verified wiring

HumanHuman operatorHuman gate
Tool
Human operator
Autonomy
Human-gated
Sends
  • directs → Supervisor
OrchestratorSupervisor
Tool
Supervisor
Autonomy
Runs autonomously
Sends
  • dispatches → Worker agent A
  • dispatches → Worker agent B
Receives
  • directs ← Human operator
  • reports ← Worker agent A
  • reports ← Worker agent B
BuilderWorker agent A
Tool
Worker agent A
Autonomy
Runs autonomously
Sends
  • reports → Supervisor
Receives
  • dispatches ← Supervisor
BuilderWorker agent B
Tool
Worker agent B
Autonomy
Runs autonomously
Sends
  • reports → Supervisor
Receives
  • dispatches ← Supervisor

How a typical Supervisor team handles a task

Typical Supervisor layout — schematic, not verified wiring

  1. Task arrives

    Human operator directs Supervisor.

  2. The orchestrator routes the work

    Supervisor dispatches build work to Worker agent A and dispatches build work to Worker agent B.

  3. The builders execute

    Worker agent A and Worker agent B build the work.

  4. Human holds the last word

    Human operator holds final approval.

Replicate a typical Supervisor setup

Typical Supervisor layout — schematic, not verified wiring

Ingredients

  • HumanHuman operator
  • OrchestratorSupervisor
  • BuilderWorker agent A
  • BuilderWorker agent B

Setup order

  1. 1.Stand up the orchestrator: Supervisor.
  2. 2.Wire Worker agent A: it receives "dispatches" from Supervisor and sends "reports" to Supervisor. Wire Worker agent B: it receives "dispatches" from Supervisor and sends "reports" to Supervisor.
  3. 3.Declare the human gate: Human operator holds final approval.

Performance metrics

Windowed metrics with provenance. [unknown] means it was not tracked — an honest hole beats an invented figure.

HumanEval Pass@1 (GPT-4, group)
89%
evidence-linked

GPT-4 multi-agent group: 89.0% vs solo 87.2% vs CoT 83.5%. Source: arXiv 2308.10848 Table 2 [evidence_linked]

as of Aug 29, 2023
Complex tool tasks completed
9
evidence-linked

9/10 complex tool-use tasks completed vs 3/10 for single ReAct agent. Source: arXiv 2308.10848 [evidence_linked]

as of Aug 29, 2023

Token economics

Cost transparency is part of the honesty architecture. [unknown] means it was not tracked — not that it is zero.

No cost metrics on record. Cost tracking is hard across runtimes; honest absence beats invented figures.

Blueprint

Operational DNA — why it works, how it was built, and how it is overseen. Not files for sale; knowledge of the design.

Why it works

Dynamic recruitment means the group is tailored to the task rather than using a static team. Peer evaluation catches errors before commitment. The coordinator synthesis role prevents individual agent biases from dominating the final answer.

How it was built

Python framework with configurable agent pool. Each agent has a persona and expertise. Coordinator uses LLM to select relevant agents from the pool for each task. Supported models: GPT-3.5, GPT-4. Open-source at github.com/OpenBMB/AgentVerse.

Oversight model

Coordinator agent manages recruitment and synthesis. Peer evaluation built into the collaboration phase. No human-in-loop described in paper evaluation.

Proof (1)

The team's shared track record — tasks, incidents, lessons, milestones. Per-entry provenance tags are always visible.

  1. ArtifactAug 21, 2023evidence-linked

    AgentVerse paper published — arXiv 2308.10848

    Dynamic group recruitment selects task-relevant specialist agents per query. Multi-agent groups outperform single agents on scientific reasoning, tabular tasks, and reading comprehension per the paper.

    https://arxiv.org/abs/2308.10848

Sign in to add a proof entry.

Sign in

Attestations (0)

Named third-party statements from people with first-hand experience. Attestations are what separates Peer-Attested from Evidence-Linked.

No attestations yet. Worked with this configuration or agent? Attest to it using the form below — attestations are named third-party statements and are what separates Peer-Attested from Evidence-Linked.

Sign in to attest to this team.

Sign in