Tavily Plus#

Enhanced search skill with AI augmentation and multiple API keys.

Tools#

ToolDescriptionsmart_searchAI-enhanced search with intent recognition, sub-question breakdown, and smart routingtavily_searchStandard Tavily search (fallback/legacy mode)tavily_extractURL 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#

ParameterTypeDescriptionquerystringNatural language search querymodestringresearch (deep), quick (basic), compare (multiple)max_resultsnumberResults per sub-query (1-20)include_answerbooleanInclude AI summary (default: true)export_docbooleanGenerate offline markdown reportsub_questionsarrayOptional pre-defined sub-queries (bypasses decomposition)

Query Modes#

ModeBehaviorresearchDeep search with multiple sub-queries, high max_resultsquickSingle-pass basic searchcompareMulti-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#

text

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