tavily-plus
# Tavily Plus
Enhanced search skill with AI augmentation and multiple API keys.
## Tools
| Tool | Description |
| ---- | ------------|
| `smart_search` | AI-enhanced search with intent recognition, sub-question breakdown, and smart routing |
| `tavily_search` | Standard Tavily search (fallback/legacy mode) |
| `tavily_extract` | URL content extraction |
## Configuration
Configure in `plugins.entries.tavily.config.webSearch`:
```json
{
"apiKey": {
"key1": "tvly-xxx1",
"key2": "tvly-xxx2"
},
"baseUrl": "https://api.tavily.com"
}
```
Or via env: `TAVILY_API_KEY`, `TAVILY_API_KEY_2`, `TAVILY_API_KEY_N`
## smart_search
AI-augmented search that performs:
1. **Intent Recognition** - Classify query type (factual, news, research, comparison, how-to, etc.)
2. **Sub-question Decomposition** - Break complex queries into atomic sub-queries
3. **Multi-key Rotation** - Rotate through configured API keys on rate limit
4. **Smart Search Execution** - Route each sub-query to optimal depth/topic
5. **Intelligent Summarization** - Synthesize findings into coherent answer
6. **Offline Document Export** - Optional markdown/PDF report generation
### Parameters
| Parameter | Type | Description |
|-----------|------|-------------|
| `query` | string | Natural language search query |
| `mode` | string | `research` (deep), `quick` (basic), `compare` (multiple) |
| `max_results` | number | Results per sub-query (1-20) |
| `include_answer` | boolean | Include AI summary (default: true) |
| `export_doc` | boolean | Generate offline markdown report |
| `sub_questions` | array | Optional pre-defined sub-queries (bypasses decomposition) |
### Query Modes
| Mode | Behavior |
|------|----------|
| `research` | Deep search with multiple sub-queries, high max_results |
| `quick` | Single-pass basic search |
| `compare` | Multi-perspective search for comparisons |
### Response Format
```json
{
"intent": "research",
"sub_questions": ["...", "..."],
"search_results": [...],
"summary": "...",
"doc_path": "/path/to/report.md" // if export_doc=true
}
```
## Key Rotation Logic
- Check `plugins.entries.tavily.config.webSearch.apiKey` (object with multiple keys)
- Fall back to env vars: `TAVILY_API_KEY`, `TAVILY_API_KEY_2`, ..., `TAVILY_API_KEY_N`
- On 429 (rate limit), automatically rotate to next key
- Log which key was used in response metadata
## Intent Classification
Supported intent types:
- `factual` - Direct answer queries (who, what, when, where)
- `how_to` - Procedural instructions
- `comparison` - A vs B analysis
- `research` - In-depth information gathering
- `news` - Current events
- `opinion` - Perspectives and opinions
## Offline Export
When `export_doc: true`:
1. Generate timestamped markdown report
2. Include: intent, sub-questions, each search result with source, AI summary
3. Save to `~/.openclaw/workspace/reports/tavily-plus-{timestamp}.md`
4. Return `doc_path` in response
## Workflow Summary
```
User Query
↓
Intent Recognition (LLM)
↓
Sub-question Decomposition (LLM)
↓
For each sub-question:
├─ Query Tavily (key rotation on 429)
└─ Collect results
↓
Intelligent Summarization (LLM)
↓
[Optional] Generate Offline Report
↓
Return Enhanced Response
```
标签
skill
ai
