Research with MAESTRO¶

MAESTRO's research system uses an agentic framework to gather information from web sources and your documents, creating comprehensive research reports through intelligent analysis and synthesis.

How Research Works¶

The Research Flow¶

1. Initial Chat Phase - You interact with the Messenger Agent to:

   • Define your research topic and requirements
   • Specify preferences (tone, format, length, audience)
   • Review and refine initial research questions
   • Add goals and constraints to the research plan

2. Research Mission Execution - Once started:

   • Chat becomes disabled during active research
   • Multiple specialized agents work autonomously
   • Progress tracked through real-time tabs
   • Notes generated and organized by topic

3. Report Generation - Upon completion:

   • Draft created from collected notes
   • Comprehensive report with citations
   • Downloadable in multiple formats

Starting a Research Session¶

Step 1: Initial Conversation¶

Begin by chatting with MAESTRO about your research needs. The Messenger Agent will:

• Help clarify your research objectives
• Generate initial exploration questions
• Capture your specific requirements

Example interactions:

"I need to research the psychological factors behind innovation adoption"
"Please focus on academic literature and use a formal tone"
"The report should be 3000-5000 words for a business audience"

Step 2: Setting Research Parameters¶

During the chat, you can specify:

• Tone: Academic, professional, conversational
• Output Format: Report structure and organization
• Length: Word count expectations
• Target Audience: Technical level and background
• Preferred Sources: Web, documents, or both
• Focus Areas: Specific aspects to emphasize

The agent adds these as goals to the research plan.

Step 3: Reviewing Initial Questions¶

The system generates initial research questions based on your topic. You can:

• Review the proposed questions
• Request modifications or additions
• Remove irrelevant questions
• Approve when satisfied

Step 4: Starting the Mission¶

Two ways to begin:

• Type "start research" or similar command
• Click the Start button (top right)

Once started, the chat interface is disabled and research begins.

Monitoring Research Progress¶

The Mission Panel Tabs¶

During research, monitor progress through four main tabs:

1. Plan Tab¶

Shows the research outline with sections and subsections:

• Outline View: Hierarchical structure
• Goal Pad: Research objectives and constraints
• Scratchpad: Agent working notes
• Thought Pad: Strategic planning notes

2. Agents Tab¶

Real-time activity monitoring:

• Activity Log: Chronological agent actions
• Agent Status: Current agent activities
• Statistics: Performance metrics
  • Total cost
  • Token usage
  • Duration
  • Success rate

Each agent entry shows:

• Agent type (ResearchAgent, PlanningAgent, etc.)
• Current action
• Cost and token consumption
• Timestamp

3. Notes Tab¶

Collected research findings:

• Research Notes: Count of notes generated
• Organization: Notes generated from web or documents
• Note Details: Each note includes:
  • Content summary
  • Source citations
  • Section assignment

Filter and search capabilities:

• By keyword
• By source type

4. Draft Tab¶

Final report generation:

• Live Preview: Report as it's being written
• Word Count: Real-time tracking
• Export Options:
  • Markdown (.md)
  • Word document (.docx)
• Edit Capability: Make final adjustments

Research Phases Explained¶

Phase 1: Initial Research¶

• Duration: 5-15 minutes
• Activities:
  • Exploring initial questions
  • Gathering background information
  • Identifying key themes
  • Building knowledge base

Phase 2: Research Planning¶

• Duration: 2-5 minutes
• Activities:
  • Creating detailed outline
  • Identifying research gaps
  • Planning investigation strategy
  • Allocating resources

Phase 3: Structured Research¶

• Duration: 10-30 minutes
• Activities:
  • Deep diving into sections
  • Gathering specific evidence
  • Cross-referencing sources
  • Building comprehensive notes

Phase 4: Writing¶

• Duration: 5-15 minutes
• Activities:
  • Synthesizing notes
  • Creating coherent narrative
  • Adding citations
  • Finalizing report

Mission Controls¶

During Research¶

Pause Button: Temporarily halt research

• Preserves current progress
• Can resume later

After Completion¶

Revise/Restart from Round: Re-run from checkpoint

• Select previous research round
• Give feedback to modify the outline and steer the research direction

Research Configuration¶

Research missions can be configured through the Settings tab or customized per mission. For a complete reference of all parameters, see the Research Parameters Reference.

Source Selection¶

Configure in Settings or per mission:

Web Search:

• Enable/disable web searching
• Configure search providers (Tavily, Jina, etc.)
• Set result limits

Local RAG (Document Search):

• Enable/disable document searching
• Select document groups

Performance Settings¶

Adjust for speed vs. thoroughness:

Research Depth:

• Number of research rounds (how many rounds to expore the topic and revise the report structure based on collected notes)
• Number of cycles per section (will gether material for a section, reflect on whats missing and can be improved, and go for another round of research)

Search Results:

How many results to examine per search query

• Initial exploration: 3-10 per source
• Main research: 5-20 per query

Concurrent Requests:

• Higher = faster but more expensive
• Lower = slower but more economical

Advanced Settings¶

Max Outline Depth: Hierarchy levels (1-4) Cycles/Section: Research iterations per section (1-3) Notes/Section: Information density (5-40) Content Limits: Maximum characters for various components to help break large context up for resource constrained local LLMs

For detailed parameter documentation, see Research Parameters Reference

Understanding the Statistics¶

Cost Tracking¶

• Total Cost: Cumulative API expenses
• Per Agent: Individual agent costs
• Token Usage: Input/output tokens

Performance Metrics¶

• Duration: Time for API calls
• Success Rate: Query success percentage
• Tool Calls: Number of searches/retrievals
• File Interactions: Documents accessed

Research Phases Chart¶

Visual breakdown showing:

• Initial Research (percentage)
• Structured Research (percentage)
• Writing (percentage)

Tips for Better Results¶

Clear Instructions¶

• Be specific about your needs
• Provide context and background
• State constraints upfront

Source Management¶

• Upload relevant documents before starting
• Create focused document groups
• Remove outdated materials

Optimal Settings¶

• Start with defaults or one of the presets
• Increase depth for complex topics
• Balance speed vs. thoroughness

Common Issues and Solutions¶

Mission Won't Start¶

• Verify AI provider configured
• Check API credits available
• Ensure at least one source enabled

Slow Progress¶

• Reduce concurrent requests
• Use faster AI models
• Limit search results

Incomplete Results¶

• Increase research depth
• Add more rounds

High Costs¶

• Use economical models for research
• Reduce search results
• Limit research rounds/cycles

Next Steps¶

1. Configure AI Providers - Set up your language models in Settings
2. Upload Documents - Build your research library in Documents
3. Start Researching - Begin with a simple topic to learn the interface
4. Optimize Settings - Adjust parameters based on your needs

For additional help, see the FAQ or visit our Community Forum.