Bootstrap Guide

How to build domain expertise with maenifold: from empty graph to institutional memory.

---

Overview

maenifold isn't just storage—it's infrastructure for building domain expertise over time. This guide walks through the full journey:

1. Seed — Initial knowledge with [[WikiLinks]]
2. Research — Workflows that explore and expand the graph
3. Specialize — Custom roles for your domain
4. Systematize — Custom workflows for your operations
5. Operate — Daily usage that compounds knowledge
6. Maintain — Sleep cycles that consolidate and prune

Important: Workflows require an LLM to drive them. The CLI provides primitives; the LLM provides intelligence.

Phase	CLI-Only	LLM Required
Seed domain	✅
Research workflows		✅
Create custom roles		✅
Create custom workflows		✅
Query/search/context	✅
Run workflows		✅
Sleep cycle		✅

---

Phase 1: Seed Your Domain

Start by writing what you already know. Every [[WikiLink]] becomes a graph node.

Tips:

• Use [[WikiLinks]] liberally—they're free and build structure
• Organize with folders: architecture/, decisions/, incidents/, runbooks/
• Link concepts that relate: [[kubernetes]] + [[deployment]] + [[rollback]]

---

Phase 2: Research & Expand

Use research workflows to systematically explore your domain. This requires an AI assistant with MCP access.

Option A: Deep Single-Agent Research

The agentic-research workflow implements HyDE, reflexion, and information gain checks:

The workflow will:

1. Establish knowledge baseline (search existing graph)
2. Generate hypothetical documents (HyDE)
3. Research external sources
4. Synthesize findings with [[WikiLinks]]
5. Check information gain; loop if insufficient

Option B: Multi-Agent Collaborative Research

The think-tank workflow orchestrates multiple agents in waves:

Waves:

1. Charter — Decompose the problem
2. Wave 1 — Parallel domain scoping
3. Wave 2 — Deep dives (nested agentic-research)
4. Wave 3 — Cross-domain synthesis
5. Wave 4 — Peer review and validation

Option C: Manual Sequential Thinking

For more control, use sequential-thinking directly:

Continue the session by passing the returned sessionId with subsequent thoughts.

---

Phase 3: Create Domain Roles

Once you have foundational knowledge, create specialist roles using constitutional AI.

Using role-creation-workflow

The workflow guides you through:

1. Define expertise — What does this role know?
2. Define principles — What does this role prioritize?
3. Define constraints — What does this role avoid?
4. Generate examples — How does this role respond?

Output: A JSON role file at assets/roles/kubernetes-architect.json

Role Structure

Adopting Roles

The role conditions how the LLM reasons over your graph—same data, different perspective.

---

Phase 4: Create Domain Workflows

With domain knowledge and a specialist role, create workflows tailored to your operations.

Using higher-order-thinking

First adopt your domain role, then design workflows:

The workflow guides meta-cognitive design:

1. Analyze the task — What does this workflow need to accomplish?
2. Identify steps — What's the logical sequence?
3. Define quality gates — How do we know each step succeeded?
4. Embed reasoning — Which steps need sequential-thinking?
5. Generate JSON — Output the workflow definition

Output: A JSON workflow file at assets/workflows/deployment-review.json

Workflow Structure

---

Phase 5: Daily Operations

Now you have:

• Knowledge graph — Growing corpus of domain knowledge
• Custom role — Specialist persona for your domain
• Custom workflows — Systematic methodologies for your tasks

Pattern: Context-First Operations

Before any task, build context:

Pattern: Workflow-Driven Tasks

For systematic work, run your custom workflows:

The workflow:

1. Builds context from graph via [[WikiLinks]]
2. Applies domain role perspective
3. Steps through systematic checks
4. Embeds sequential-thinking for complex reasoning
5. Writes findings back to graph

Pattern: Capture As You Go

Every insight should become a memory:

The graph grows with every operation. Future sessions benefit from past work.

---

Phase 6: Maintenance

Knowledge isn't static. Memories decay without use. Sleep cycles consolidate and prune.

Sleep Cycle

Run periodically (daily recommended):

Three specialists run in parallel; decay is handled within the memory-cycle workflow itself:

1. Consolidation — Replays high-significance episodic memories, promotes to semantic
2. Repair — Normalizes WikiLink variants, cleans orphaned concepts
3. Epistemic — Reviews assumptions, validates or invalidates based on evidence

What Happens

Memory Type	Grace Period	Half-Life	Fate
Episodic (thinking/)	7 days	30 days	Consolidated or decayed
Semantic (research/, decisions/)	28 days	30 days	Stable if accessed
Immortal (validated assumptions)	∞	∞	Permanent

Dream Synthesis

During consolidation, FindSimilarConcepts discovers novel connections across domains:

These insights surface during sleep—connections you didn't explicitly make.

---

The Compound Effect

Every [[WikiLink]] compounds. Every workflow run teaches. Every sleep cycle refines.

The graph becomes your second brain for the domain—one that remembers, forgets appropriately, and surfaces what matters.

---

Quick Reference

CLI Primitives

Key Workflows

Workflow	Purpose
agentic-research	Deep single-agent research with HyDE and reflexion
think-tank	Multi-agent collaborative research in waves
role-creation-workflow	Create domain-specific roles
higher-order-thinking	Design custom workflows
workflow-dispatch	Auto-select optimal methodology
memory-cycle	Consolidation, decay, repair, epistemic maintenance

MCP Configuration

---

The graph grows smarter over time. So does your AI assistant.