Skip to content

AgentEmulator v1.0.0

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 08 Oct 07:18
· 4 commits to main since this release
v1.0.0
1950acc
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

AgentEmulator v1.0.0

The first public release of AgentEmulator, an emulation and experimentation platform for trustworthy AI agent infrastructure, built by HuangLab (Sun Yat-sen University) on top of BlockEmulator-X.

Highlights

  • Trace-driven behavior replay. Describe agent actions (join / pay / leave) and plain transfers in a JSONL trace; the agentSupervisor module compiles the trace into a transaction dataset and replays it on a fresh BlockEmulator-X blockchain (PBFT, default 4 shards × 4 nodes).
  • Deterministic identity. DIDs are derived from the configured seed and agent_id — no DIDs in traces — and agent lifecycle state is tracked in agent_registry.json.
  • Real settlement. pay actions execute as actual on-chain balance transfers; join / leave are recorded as DID register / revoke contract calls (ToDo).
  • Experiment data. Per-agent ledger CSVs, an action-to-transaction mapping (request_id → tx hashes), and chain relay statistics for payment confirmation and latency analysis.
  • Visualization. Publication-styled balance figures and a HTML gallery, generated automatically after each run.
  • Reproducibility. The same trace, seed, and code version produce a byte-identical transaction dataset and deterministic ledger ordering.

The default demo trace (100 agents, 10,000 transactions) and configuration are included, along with run_agentemu.sh / run_agentemu.bat launch scripts.

Quick start

Requires Go ≥ 1.25 and Python ≥ 3.8 with matplotlib / numpy / pandas.

git clone https://github.com/HuangLab-SYSU/agent-emulator.git
cd agent-emulator
go build ./...
bash run_agentemu.sh          # Windows: run_agentemu.bat

The script builds the project, runs the experiment on the default trace, generates the figures, and opens the HTML gallery.

Distribution

This is a source-only release — no prebuilt binaries are shipped. Clone the repository and run from source as shown in Quick start: the agentemu entry point compiles the chain binaries at start-up, which requires a local Go toolchain (≥ 1.25).

Citation

If you use AgentEmulator in your research, please cite the accompanying paper (DOI: 10.13140/RG.2.2.23825.60000).

AgentEmulator v1.0.0（中文说明）

AgentEmulator 的首个公开发布版本。AgentEmulator 是由中山大学 HuangLab 基于区块链仿真平台 BlockEmulator-X 构建的可信 AI Agent 基础设施仿真与实验平台。

版本要点

  • Trace 驱动行为回放。 在 JSONL trace 中描述 agent 行为（join / pay / leave）与普通转账，agentSupervisor 模块将其编译为交易数据集，并在全新的 BlockEmulator-X 区块链（PBFT，默认 4 分片 × 每分片 4 节点）上回放。
  • 确定性身份。 DID 由配置的 seed 与 agent_id 派生（trace 中无需填写 DID），agent 生命周期状态记录于 agent_registry.json。
  • 真实结算。 pay 动作以真实链上余额转账执行；join / leave 记录为 DID register / revoke 合约调用（ToDo）。
  • 实验数据。 输出 per-agent 账本 CSV、动作到交易的映射（request_id → 交易哈希）以及链上 relay 统计，支持支付确认与延迟分析。
  • 可视化。 自动生成论文级余额图表，并构建 HTML gallery。
  • 可复现。 相同 trace、seed 与代码版本生成字节级一致的交易数据集与确定性的账本排序。

仓库附带默认演示 trace（100 个 agent、10000 笔交易）以及 run_agentemu.sh / run_agentemu.bat 启动脚本。

快速开始

需要 Go ≥ 1.25 与 Python ≥ 3.8（含 matplotlib / numpy / pandas）。

git clone https://github.com/HuangLab-SYSU/agent-emulator.git
cd agent-emulator
go build ./...
bash run_agentemu.sh          # Windows: run_agentemu.bat

脚本会自动构建、在默认 trace 上运行实验、生成图表并打开 HTML gallery。

分发方式

本 release 仅提供源码，不附带预构建二进制。请按「快速开始」克隆仓库并从源码运行：agentemu 入口会在启动时编译链上二进制，因此需要本地 Go 工具链（≥ 1.25）。

引用

如果在研究中使用了 AgentEmulator，请引用随附论文（DOI: 10.13140/RG.2.2.23825.60000）。

Assets 2
Loading