Skip to content

sii_deep_research - SII Deep Research

Overview

The sii_deep_research tool conducts comprehensive research using the Engineer Deep Research API. It features auto-start research sessions with milestone tracking, real-time phase analysis, session history, phase-specific information extraction, and service health monitoring. Supports both Chinese and English research outputs. Only available when authenticated with SII.

Tool Name

  • Internal Name: sii_deep_research
  • Display Name: SII Deep Research
  • Icon: Search

Prerequisites

To use this tool, you need:

  1. SII Account: A valid SII account is required
  2. Authentication: Must be authenticated with SII
  3. Environment Variables:
    • SII_USERNAME: SII username
    • SII_PASSWORD: SII password

Parameters

Required Parameters

ParameterTypeDescription
actionstringAction to perform: "start", "health", "history", "phase_analysis", "extract_phase_info", "monitor", "milestone_summary", "save_results".

Optional Parameters

ParameterTypeDescription
research_questionstringThe research question to investigate thoroughly. Required for "start" action.
languagestringLanguage for research output: "zh" (Chinese) or "en" (English). Default: "zh".
session_idstringSession ID for operations requiring an existing session. Required for: "history", "phase_analysis", "extract_phase_info", "save_results" actions.
target_phasestringTarget phase for "extract_phase_info" action: "planning", "web_search", "browsing", "writing", "completion", "session_end".
research_paramsobjectResearch configuration parameters.
research_params.max_iterationsnumberMaximum number of research iterations (default: 2).
research_params.web_search_enabledbooleanEnable web search during research (default: true).
research_params.cognition_search_enabledbooleanEnable cognition database search (default: true).
metadataobjectCustom metadata for the research session.
metadata.created_bystringCreator of the research session.
metadata.descriptionstringDescription of the research session.
limitnumberLimit for history retrieval (default: 100, max: 1000).

Actions

1. start - Start Research

Creates and auto-starts a new research session.

json
{
  "action": "start",
  "research_question": "What are the latest developments in quantum computing?",
  "language": "en",
  "research_params": {
    "max_iterations": 3,
    "web_search_enabled": true,
    "cognition_search_enabled": true
  }
}

2. health - Check Service Health

Checks the health status of the deep research service.

json
{
  "action": "health"
}

3. history - Get Session History

Retrieves research session history.

json
{
  "action": "history",
  "session_id": "session_123",
  "limit": 50
}

4. phase_analysis - Analyze Current Phase

Analyzes the current phase of a research session.

json
{
  "action": "phase_analysis",
  "session_id": "session_123"
}

5. extract_phase_info - Extract Phase Information

Extracts information from a specific research phase.

json
{
  "action": "extract_phase_info",
  "session_id": "session_123",
  "target_phase": "web_search"
}

6. monitor - Monitor Research Progress

Monitors research progress with milestone tracking.

json
{
  "action": "monitor",
  "session_id": "session_123"
}

7. milestone_summary - Get Milestone Summary

Gets a quick milestone summary with content previews.

json
{
  "action": "milestone_summary",
  "session_id": "session_123"
}

8. save_results - Save Research Results

Saves complete research results to a local file when research is completed.

json
{
  "action": "save_results",
  "session_id": "session_123"
}

Research Phases

The deep research process goes through several phases:

  1. planning: Initial research planning and strategy
  2. web_search: Web search for relevant information
  3. browsing: Browsing and analyzing web content
  4. writing: Writing and synthesizing research findings
  5. completion: Finalizing research results
  6. session_end: Research session completed

Features

  1. Comprehensive Research:

    • Multi-phase research process
    • Web search integration
    • Cognition database search
    • Iterative refinement
  2. Real-time Tracking:

    • Phase progress monitoring
    • Milestone tracking
    • Status updates
    • Progress indicators
  3. Flexible Configuration:

    • Adjustable iteration count
    • Enable/disable search sources
    • Language selection
    • Custom metadata
  4. Result Management:

    • Session history
    • Phase-specific extraction
    • Complete result export
    • Milestone summaries

Usage Examples

Start Basic Research

json
{
  "action": "start",
  "research_question": "Explain the transformer architecture in deep learning",
  "language": "en"
}

Start Advanced Research

json
{
  "action": "start",
  "research_question": "Compare different approaches to few-shot learning",
  "language": "en",
  "research_params": {
    "max_iterations": 5,
    "web_search_enabled": true,
    "cognition_search_enabled": true
  },
  "metadata": {
    "created_by": "researcher_name",
    "description": "Comparative analysis of few-shot learning methods"
  }
}

Monitor Research Progress

json
{
  "action": "monitor",
  "session_id": "session_123"
}

Save Research Results

json
{
  "action": "save_results",
  "session_id": "session_123"
}

Return Value

For "start" Action

typescript
{
  session_id: string;
  status: string;
  message: string;
}

For "phase_analysis" Action

typescript
{
  session_id: string;
  current_phase: string;
  phase_progress: number;
  status: string;
  milestones: Array<{
    phase: string;
    timestamp: string;
    status: string;
  }>;
}

For "save_results" Action

typescript
{
  file_path: string;
  content: string;
  summary: string;
}

Error Handling

Possible error scenarios:

  1. Authentication Errors:

    • Not authenticated with SII
    • Invalid credentials
    • Expired session
  2. Parameter Errors:

    • Missing required parameters
    • Invalid action type
    • Invalid session ID
  3. Research Errors:

    • Research session not found
    • Research still in progress
    • Research failed
  4. Network Errors:

    • Connection timeout
    • Service unavailable
    • Network interruption

Best Practices

  1. Research Question:

    • Be specific and clear
    • Provide sufficient context
    • Avoid overly broad questions
  2. Configuration:

    • Adjust iterations based on complexity
    • Enable relevant search sources
    • Choose appropriate language
  3. Monitoring:

    • Check progress periodically
    • Don't poll too frequently
    • Wait for completion before saving
  4. Result Management:

    • Save results when research completes
    • Review milestone summaries
    • Extract phase-specific information as needed

Research Duration

Deep research typically takes 5-10 minutes to complete, depending on:

  • Question complexity
  • Number of iterations
  • Search sources enabled
  • Network conditions

Notes

  1. Authentication Required: This tool requires SII authentication
  2. Long Running: Research sessions can take several minutes
  3. Resource Intensive: May consume significant API resources
  4. Language Support: Supports both Chinese and English output
  5. Async Operation: Research runs asynchronously
  • sii_web_search: Search the web using SII
  • sii_cognitions: Search SII cognitions database
  • sii_hybrid_search: Combined web and cognition search
  • write_file: Save research results to files

Released under the MIT License.