Full-Code · Orchestration Frameworksactive

AgentVerse

Type: full-code · Vendor: OpenBMB / Tsinghua · Language: Python · License: Apache-2.0 · Status: active · Status in practice: emerging

Links: homepage repo

Multi-agent framework with two distinct modes: task-solving (collaborative agents work toward a shared goal) and simulation (autonomous agents interact in an environment to study emergent social behaviour).

Description. AgentVerse is OpenBMB's multi-agent framework with a deliberate dual-mode design. In task-solving mode, several role-specialised agents coordinate to solve a single user task (similar in spirit to AutoGen GroupChat). In simulation mode, autonomous agents interact in a configured environment (e.g. a simulated software-development office, a research-paper writing committee, a virtual classroom) to study emergent collective behaviour. The framework provides shared environment abstractions (agents, environments, rules, recorders) that both modes plug into.

Agent loop shape. Environment-loop. The environment ticks; on each tick, agents perceive shared state, act (speak, use tools, move), and the environment updates. Task-solving mode runs a small fixed agent team until task termination. Simulation mode runs N autonomous agents indefinitely with no shared goal, logging interactions for offline analysis.

Primary use cases

multi-agent task-solving with role specialisation
social-simulation research (agent societies, emergent behaviour)
education: configurable multi-agent classrooms and debate environments
studying communication protocols between non-cooperative agents

flowchart TD CONFIG[Environment config] --> MODE{Mode} MODE -->|task-solving| TASK[Task team<br/>role-specialised agents] MODE -->|simulation| SIM[Agent society<br/>N autonomous agents] TASK --> ENV[(Environment<br/>shared state)] SIM --> ENV ENV --> RULES[Rules / order] RULES --> TICK[Tick: agents perceive → act → state updates] TICK --> ENV ENV --> RECORDER[Recorder<br/>log interactions] TASK -.terminate on goal.-> END[Result] SIM -.run indefinitely.-> RECORDER

Key concepts

Task-solving mode → group-chat-manager — Collaborative multi-agent coordination toward a shared goal.
Simulation mode → actor-model-agents — Autonomous agent society for studying emergent behaviour.
Environment / agent / rule abstraction → blackboard — Pluggable triad that both modes share.