企业级AI的空白
企业Java缺少可视化AI编排中间件。团队要么用Python/Node.js工具制作原型后为生产重写，要么使用Spring AI或LangChain4j等库从头构建一切。
EDDI填补这一空白
EDDI是可部署的中间件平台，而不是库。它提供团队所需的一切：
- 可视化管理界面 — EDDI管理器用于构建和监控代理
- 配置即代码 — 代理逻辑是JSON，不是编译的代码
- 42个MCP工具 — 通过模型上下文协议实现完整的AI原生控制
- 企业安全 — OIDC、密钥库、审计跟踪、无eval()
- 生产基础设施 — REST API、对话状态管理、Prometheus指标
- 水平扩展 — NATS JetStream用于分布式架构
谁应该使用EDDI？
- 企业团队需要生产就绪的AI编排平台
- 提示词工程师想要在无需重新部署的情况下迭代
- 受监管行业需要审计跟踪和欧盟AI法案合规
- 平台团队为多个部门构建内部AI服务
EDDI 与典型代理框架的比较
| 维度 | Python/Node 框架 | EDDI |
|---|---|---|
| Concurrency | GIL or single-threaded event loop | Java 25 Virtual Threads — true OS-level parallelism |
| Agent Logic | Embedded in application code | Versioned JSON configs — update behavior without redeployment |
| Security Model | Relies on sandboxed code execution | No dynamic code execution; envelope-encrypted vault, SSRF protection |
| Compliance | Requires custom implementation | GDPR, HIPAA, EU AI Act infrastructure built-in |
| Audit Trail | Application-level logging | HMAC-SHA256 immutable ledger with cryptographic agent signing |
| Deployment | pip/npm + manual infrastructure | One-command Docker install, Kubernetes/OpenShift-ready |
支持 12 个 LLM 提供商
连接任何主要 LLM 提供商 — 或通过任何兼容 OpenAI 的端点使用你自己的。
| 类别 | 提供商 |
|---|---|
| 云端 API | OpenAI · Anthropic Claude · Google Gemini · Mistral AI |
| 企业云 | Azure OpenAI · Amazon Bedrock · Oracle GenAI · Google Vertex AI |
| 自托管 | Ollama · Jlama · Hugging Face |
| 兼容 | 任何 OpenAI 兼容端点（DeepSeek、Cohere 等）通过 baseUrl |
每位CIO应该问的8个问题
在评估AI代理编排平台时，这些问题 that separate production-grade infrastructure from fragile prototypes:
Architecture
Does the platform execute user-supplied code at runtime?
EDDI: No. EDDI uses declarative JSON configuration only.
Performance
How does the platform handle thousands of concurrent agent conversations?
EDDI: Java 25 virtual threads provide true OS-level parallelism.
Integration
Can agents consume external tools through open standards?
EDDI: 42 MCP tools, A2A protocol, OpenAPI 3.1, OAuth 2.0/OIDC.
Data Privacy
How are data subject rights implemented?
EDDI: Unified REST API cascades across all 5 data stores. 15+ frameworks.
Security
What is the platform's architectural security posture?
EDDI: No dynamic code execution. OIDC/Keycloak, AES-256-GCM vault, HMAC-SHA256 audit trails.
Compliance
Does the platform provide immutable audit trails?
EDDI: HMAC-SHA256 tamper-evident ledger with full pipeline tracing.
Operations
Can non-developers use the platform?
EDDI: EDDI Manager: production-ready React UI with visual agent building.
Portability
Can it run on-premises and air-gapped?
EDDI: Docker-native. 12 LLM providers. Full air-gap via Ollama.
总拥有成本：自建 vs. 部署
使用AI库的隐藏成本并非在于库本身 — it's the invisible infrastructure teams must build around it:
使用库自建
- Custom REST API layer (2–4 weeks)
- Authentication & RBAC system (2–3 weeks)
- Conversation state persistence (1–2 weeks)
- Audit trail & compliance logging (2–4 weeks)
- Management UI for non-developers (4–8 weeks)
- Secret management integration (1–2 weeks)
- Horizontal scaling & coordination (2–4 weeks)
- Ongoing maintenance & security patching
部署EDDI
- One-command install (5 minutes)
- All of the above included out of the box
- Team focuses on business logic, not infrastructure
- Maintained by an 18-year-old open-source project
商业案例
EDDI的价值在于团队无需自行构建的部分.
对于受监管的行业来说，成本等式更加清晰：替代方案是一个覆盖15+个监管框架的自定义合规实施。