geepers-orchestrate

通过 dr.eamer.dev 编排 API 运行多代理 Dream Cascade（分层 3 层综合）或 Dream Swarm（并行多域搜索）工作流程。当任务受益于多个并行或分层工作的专用代理时使用。

# Dreamer Orchestrate

Run multi-agent workflows through `https://api.dr.eamer.dev`.

## Authentication

```bash
export DREAMER_API_KEY=your_key_here
```

## Endpoints

### Dream Swarm — Parallel Search
```
POST https://api.dr.eamer.dev/v1/orchestrate/swarm
Body:
{
  "query": "What are the most effective treatments for Type 2 diabetes?",
  "sources": ["pubmed", "semantic_scholar", "wikipedia"],
  "num_agents": 5
}
```
Runs multiple agents simultaneously across data sources and synthesizes results.

### Dream Cascade — Hierarchical Synthesis
```
POST https://api.dr.eamer.dev/v1/orchestrate/cascade
Body:
{
  "task": "Analyze the current state of quantum computing hardware",
  "num_agents": 8,
  "provider": "anthropic"
}
```
Three-tier workflow: Belter agents gather raw data → Drummer agents synthesize domains → Camina produces executive summary.

## Response Format

Both endpoints return:
```json
{
  "result": "Synthesized answer...",
  "sources": [...],
  "agent_count": 5,
  "duration_ms": 12450
}
```

## When to Use
- Complex research questions requiring multiple perspectives
- Cross-domain synthesis that would take multiple sequential queries
- Long-horizon analysis where parallelism saves time

## Don't Use When
- Simple single-source queries (use dreamer-data instead)
- You need fine-grained control over individual agent behavior
- Latency is critical (orchestration takes 10-60 seconds)

---

## 中文说明

# Dreamer Orchestrate

通过 `https://api.dr.eamer.dev` 运行多代理工作流。

## 身份验证

```bash
export DREAMER_API_KEY=your_key_here
```

## 端点

### Dream Swarm —— 并行搜索
```
POST https://api.dr.eamer.dev/v1/orchestrate/swarm
Body:
{
  "query": "What are the most effective treatments for Type 2 diabetes?",
  "sources": ["pubmed", "semantic_scholar", "wikipedia"],
  "num_agents": 5
}
```
跨数据源同时运行多个代理并综合结果。

### Dream Cascade —— 分层综合
```
POST https://api.dr.eamer.dev/v1/orchestrate/cascade
Body:
{
  "task": "Analyze the current state of quantum computing hardware",
  "num_agents": 8,
  "provider": "anthropic"
}
```
三层工作流：Belter 代理收集原始数据 → Drummer 代理综合各领域 → Camina 生成执行摘要。

## 响应格式

两个端点都返回：
```json
{
  "result": "Synthesized answer...",
  "sources": [...],
  "agent_count": 5,
  "duration_ms": 12450
}
```

## 何时使用
- 需要多种视角的复杂研究问题
- 本需多次顺序查询才能完成的跨领域综合
- 并行能节省时间的长程分析

## 何时不要使用
- 简单的单源查询（改用 dreamer-data）
- 你需要对单个代理行为进行细粒度控制
- 延迟至关重要时（编排需要 10-60 秒）