Skip to main content

Knowledge Building: Research Workflows + Obsidian Navigation

Use AI research workflows to build rich knowledge graphs, then navigate visually in Obsidian

Prerequisites

  • Claude Code installed (or compatible MCP client)
  • Obsidian installed (optional, for visual navigation)
  • Topic to research (e.g., FinOps Framework, system architecture)

Setup

1.

Install maenifold:

npm install -g @ma-collective/maenifold
2.

Configure Claude Code with maenifold MCP server (see Quick Start)

3.

Optional: Set MAENIFOLD_ROOT to Obsidian vault for visual navigation

4.

Verify with: "Use agentic-research workflow to research [topic]"

Walkthrough Examples

Example 1: Single-Agent Deep Research (agentic-research)

Scenario: Build comprehensive knowledge about FinOps Framework to prepare for UC2.

Claude Code Prompt:

"Use agentic-research workflow to research the FinOps Framework"

What AI Does (11-step workflow):

  1. Research Initiation: Creates research plan with [[coverage-vector]]
  2. Knowledge Baseline: Searches memory for existing [[FinOps]] knowledge
  3. HyDE Query Expansion: Generates hypothetical documents for semantic search
  4. Information Gathering: Executes web research, creates [[cost-optimization]] [[commitment-strategy]] concepts
  5. Topic Discovery: Identifies emergent themes like [[rate-optimization]]
  6. Multi-Perspective Analysis: Analyzes from analyst, practitioner, CFO viewpoints
  7. Synthesis & Reflexion: Six thinking hats quality review
  8. Knowledge Integration: Creates memory://research/finops-framework/ with [[WikiLinks]]
  9. Information Gain: Validates 20%+ new knowledge gained
  10. Research Deliverable: Comprehensive report with findings
  11. Quality Validation: Scores 7/10+ research quality

Output:

  • 10+ memory files at memory://research/finops-framework/
  • 50+ [[concepts]] in knowledge graph
  • Ready for UC2 domain extensibility

Why This Matters: You've seeded the graph with domain knowledge that AI will use to create custom roles.

Example 2: Multi-Agent Think Tank (think-tank)

Scenario: Complex research requiring parallel agent investigation.

Claude Code Prompt:

"Use think-tank workflow to research cloud financial optimization"

What AI Does (4 waves, 15-20 agents):

Wave 1: Domain Scoping (4 parallel agents)

  • Agent 1: Maps [[knowledge-landscape]] and [[domains]]
  • Agent 2: Scans [[academic-literature]] and [[seminal-works]]
  • Agent 3: Identifies [[emerging-patterns]] and [[trends]]
  • Agent 4: Finds [[knowledge-gaps]] and [[opportunities]]

Wave 2: Deep Dive (3-5 parallel agents per domain)

  • Each uses full agentic-research workflow
  • 20+ [[WikiLinks]] per domain
  • Achieves 30%+ information gain

Wave 3: Cross-Domain Synthesis (4 parallel agents)

  • Finds [[interdisciplinary-connections]]
  • Identifies [[emergent-patterns]]
  • Reconciles contradictions
  • Generates [[novel-insights]]

Wave 4: Peer Review (4 parallel agents)

  • Validates methodology (7/10+ score)
  • Assesses evidence quality
  • Verifies logic chains
  • Confirms 80%+ coverage

Output:

  • 40+ memory files with comprehensive domain knowledge
  • 100+ [[concepts]] in interconnected graph
  • Multi-perspective synthesis ready for domain work

Why This Matters: Parallel research accelerates knowledge building for complex domains.

Example 3: Obsidian Visual Navigation

Scenario: After research, navigate your knowledge graph visually.

Steps:

  1. Open memory folder as Obsidian vault
  2. View graph visualization: [[FinOps-Framework]] connects to [[cost-optimization]], [[commitment-strategy]], [[ROI]]
  3. Click [[WikiLinks]] to navigate between research files
  4. Use backlinks panel to see all mentions of [[Azure]]
  5. Search across entire research corpus instantly

Why Obsidian?

  • Visual graph view shows concept relationships
  • Bidirectional links surface unexpected connections
  • Markdown-native, no lock-in
  • Works alongside maenifold seamlessly

Why This Matters: Visual navigation accelerates understanding without leaving your workspace.

Built-In Workflows

Two research workflows bundled with maenifold:

agentic-research

Single-agent, 11 steps, HyDE-enhanced

/assets/workflows/agentic-research.json

think-tank

Multi-agent waves, 15-20 agents, parallel execution

/assets/workflows/think-tank.json

Code Sample

# Single-agent research via CLI
maenifold --tool Workflow --payload '{
  "workflowId": "agentic-research",
  "response": "Research the FinOps Framework"
}'

# Multi-agent research (requires Claude Code with agent orchestration)
# Use think-tank workflow via Claude Code

# Optional: Visual navigation with Obsidian
export MAENIFOLD_ROOT="~/Documents/ObsidianVault"

Common Pitfalls

⚠️
Workflow choice: Use agentic-research for focused research, think-tank for complex multi-domain topics
⚠️
Concept density: Aim for 10-50+ [[WikiLinks]] per research file for rich graph connections
⚠️
Sync after research: Run Sync tool to update graph database after creating memories
⚠️
Obsidian optional: Graph works without Obsidian—it just provides visual navigation layer
⚠️
WikiLink syntax: Use [[concept-name]] with hyphens for multi-word concepts, not spaces

Next Steps