经 AI Skill Hub 精选评估,AgentScope Go 获评「强烈推荐」。这款Agent工作流在功能完整性、社区活跃度和易用性方面表现出色,AI 评分 8.0 分,适合有一定技术背景的用户使用。
AgentScope Go 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
AgentScope Go 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:go install(推荐) go install github.com/AlanFokCo/agentscope-go@latest # 方式二:从源码编译 git clone https://github.com/AlanFokCo/agentscope-go cd agentscope-go go build -o agentscope-go . # 方式三:下载预编译二进制 # 访问 Releases 页面下载对应平台二进制文件 # https://github.com/AlanFokCo/agentscope-go/releases
# 查看帮助 agentscope-go --help # 基本运行 agentscope-go [options] <input> # 详细使用说明请查阅文档 # https://github.com/AlanFokCo/agentscope-go
# agentscope-go 配置说明 # 查看配置选项 agentscope-go --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export AGENTSCOPE_GO_CONFIG="/path/to/config.yml"
<p align="center"> <img src="https://img.alicdn.com/imgextra/i1/O1CN01nTg6w21NqT5qFKH1u_!!6000000001621-55-tps-550-550.svg" alt="AgentScope Logo" width="200" /> </p>
<p align="center"> <a href="https://github.com/agentscope-ai/agentscope">🐍 Python</a> | <a href="https://github.com/agentscope-ai/agentscope-java">☕ Java</a> | <a href="README.es-ES.md">Español</a> </p>
<p align="center"> <img src="https://img.shields.io/badge/license-Apache--2.0-blue" alt="License" /> <img src="https://img.shields.io/badge/Go-1.25%2B-00ADD8?logo=go" alt="Go 1.25+" /> <a href="https://pkg.go.dev/github.com/alanfokco/agentscope-go/v2/pkg/agentscope"><img src="https://pkg.go.dev/badge/github.com/alanfokco/agentscope-go/v2/pkg/agentscope.svg" alt="Go Reference" /></a> </p>
---
AgentScope Go is the Go implementation of the AgentScope multi-agent LLM framework. It provides Go-idiomatic APIs — interfaces, context.Context, explicit error returns, functional options — while delivering capabilities that go beyond the Python project.
---
These sections describe capabilities provided by this Go repository. They are not a version-by-version comparison with the Python project.
Requirements: Go 1.25+
go get github.com/alanfokco/agentscope-go/v2/pkg/agentscope
export DASHSCOPE_API_KEY=sk-... # or ANTHROPIC_API_KEY / OPENAI_API_KEY
go run ./examples/agent_v2
54 examples in examples/. Run any with go run ./examples/<name>.
| Example | Description |
|---|---|
| **Agent Basics** | |
simple | Minimal agent + single chat call |
agent_v2 | UnifiedAgent with native API tool calling |
streaming | Real-time streaming via ReplyStream + event channel |
react_tool | UnifiedAgent with custom FunctionTool |
react_builtin_tools | UnifiedAgent with the enhanced built-in toolkit (bash, read, write, edit, multiedit, applypatch, glob, grep) |
console | Interactive terminal chat with an agent (streamed rendering + tool-call confirmation) |
dingtalk_channel | Connect an agent to DingTalk (Stream SDK inbound, webhook replies, text-mode confirmations) |
| **Model API** | |
model_call | Raw model API: streaming + two-round tool calling + structured output |
structured_output | Force JSON Schema-compliant output via GenerateStructuredOutput |
multi_provider | Model card queries + 9-provider switching |
multimodal | Image input via URL and Base64 DataBlock |
multiagent | Multi-agent conversation with moderator summary |
multiagent_multimodal | Multi-agent + shared image input |
openai_response | OpenAI Responses API (call + tools + structured output) |
| **Infrastructure** | |
middleware | Custom logging middleware (model call + tool execution hooks) |
permission | Permission engine: Explore / Default / Bypass modes |
tracing | OpenTelemetry-style tracing with nested spans |
agent_loop | v3 agent loop with MetricsHook and InMemoryProvider |
embedding | Text embedding + cosine similarity matrix |
long_term_memory | Cross-session memory middleware (3 modes) |
agentic_memory | File-based memory: FileStore JSON Lines persistence + MEMORY.md agentic middleware |
rag_react | RAG with in-memory index + knowledge base |
| **Multi-Agent & Orchestration** | |
pipeline_multi_agent | Pipeline + MsgHub orchestration |
agent_team | Leader/Worker team with message routing |
mcp | MCP client: tool discovery + remote execution |
a2a_http | Agent-to-Agent over HTTP |
| **Go Runtime** | |
replay | Record LLM calls, replay in CI without API costs |
agent_pool | Fan-out agent pool with backpressure |
hotreload | Zero-downtime config updates with typed Reloader[T] |
wasm_sandbox | WASM tool sandbox with memory/time limits |
grpc_a2a | TCP agent mesh with bidirectional streaming |
bench | Agent load testing with P50/P95/P99 latency |
hub_install | Browse and install MCP servers/skills from hub |
skill_partitions | Per-agent skill partitions: .seed template, equip-once, migration, purge |
workspace_sharing | Session-workspace sharing + read-only artifact endpoints (list_dir/read_file) |
access_control | Resource sharing across users/groups/orgs |
document_parser | Parse PDF/Word/Excel/PPT into RAG chunks |
audit_logging | Sandbox policy enforcement + structured audit trail |
guardrail | Output content filtering with block/redact/warn actions |
eval_harness | Replay-based agent evaluation with scorers |
replayview | Terminal viewer for RunJSONL run logs (step through events) |
rundiff | Align two RunJSONL run logs and print where they diverge |
spend_cap | USD/CNY spend cap with CostTrackerMiddleware |
| **Deployment** | |
agent_service | HTTP Agent Service (REST + SSE streaming) |
webui | Web UI Studio with streaming chat, tool visualization, HITL |
scheduled_task | One-shot and recurring task scheduling |
realtime_echo | Realtime streaming interface demo |
| **Edge & IoT** | |
edge_offline | ConnectivityAwareModel — automatic cloud/local fallback |
edge_sensor | SensorMiddleware + Watchdog for physical sensors |
edge_serial_robot | DeviceTool with serial robot arm control |
edge_fleet | Multi-agent PubSub coordination across devices |
| **Multi-Agent Games** | |
werewolves | Multi-agent Werewolves game with role-based behavior |
| **Tracing** | |
tracing_otlp | OTLP tracing setup pattern (no OTel SDK dependency) |
| **Kubernetes** | |
k8s_workspace | K8s workspace sandboxing + cluster read-only tools |
---
Zero-downtime configuration updates with typed generics. File changes are detected by polling; the new config is atomically swapped in.
type AgentCfg struct {
Model string `json:"model"`
Temperature float64 `json:"temperature"`
MaxTokens int `json:"max_tokens"`
}
watcher := hotreload.NewWatcher(hotreload.WatcherConfig{PollInterval: 2 * time.Second})
reloader, _ := hotreload.NewReloader[AgentCfg](watcher, "config/agent.json",
hotreload.WithOnChange(func(old, new_ *AgentCfg) {
log.Printf("model changed: %s -> %s", old.Model, new_.Model)
}),
)
watcher.Start(ctx)
// Always reads the latest config — no restart needed
cfg := reloader.Get()
| Protocol | Description |
|---|---|
| **MCP** | MCP client over Stdio and HTTP JSON-RPC, with automatic tool discovery. There is no SSE/streamable-HTTP transport (see STABILITY.md → *Deliberately not ported*) |
| **A2A HTTP** | Agent-to-Agent over HTTP via A2AAgent + HTTPClient |
| **A2A TCP** | Newline-delimited JSON transport; the a2a/grpc package does not implement the gRPC wire protocol |
| **AG-UI** | Agent service protocol for frontend integration |
| **Agent Teams** | Leader/Worker coordination with cross-session HITL event projection |
| **Pipeline & MsgHub** | Sequential Then/If combinators + multi-agent message routing |
AgentScope Go是一个有前途的开源AI工作流框架
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ Apache 2.0 — 宽松开源协议,可商用,需保留版权声明和 NOTICE 文件,含专利授权条款。
AI Skill Hub 点评:AgentScope Go 的核心功能完整,质量优秀。对于自动化工程师和运维人员来说,这是一个值得纳入个人工具库的选择。建议先在非生产环境试用,再逐步推广。
| 原始名称 | agentscope-go |
| 原始描述 | 开源AI工作流:AgentScope Golang: Agent-Oriented Programming for Building LLM Applications。⭐22 · Go |
| Topics | AIagentchatbotGo |
| GitHub | https://github.com/AlanFokCo/agentscope-go |
| License | Apache-2.0 |
| 语言 | Go |
收录时间:2026-07-02 · 更新时间:2026-07-03 · License:Apache-2.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端