Back to Docs

Memory Types

5 Types of Context-Aware Memory

"Context that persists, decisions that compound, insights that accumulate"

3-Layer Context System

Memory is automatically loaded and injected at three levels to ensure agents always have the context they need.

1

Layer 1: Keywords

Auto-load context with triggers like "my research", "연구 진행"

Example:User says "my research" → Project Context automatically loaded
2

Layer 2: Task Tool

Auto-injects context to agents via Task tool delegation

Example:Agent receives Project Context in system prompt automatically
3

Layer 3: CLI

Explicit memory commands via CLI

Example:/diverga:memory context → View current memory state

Five Memory Types

Each memory type serves a specific purpose in maintaining research continuity and context.

Project Context

Research question, paradigm, methodology, and theoretical framework

Auto-loaded

Storage:

.research/project-state.yaml

Examples:

  • Research question and objectives
  • Selected paradigm (positivist, interpretivist, etc.)
  • Chosen theoretical framework with T-Score
  • Research design approach

Session Memory

Current session state, active tasks, and in-progress work

Auto-loaded

Storage:

.research/sessions/session-{id}.yaml

Examples:

  • Active agent tasks and status
  • Current analysis stage
  • In-progress calculations
  • Session-specific variables

Decision Log

All human checkpoint decisions with timestamps and rationale

Auto-loaded

Storage:

.research/decision-log.yaml

Examples:

  • Checkpoint approvals/rejections
  • Selected alternatives with T-Scores
  • Decision timestamps
  • Rationale and justifications

Research Notes

Accumulated insights, findings, and observations

Storage:

.research/notes/

Examples:

  • Key findings from literature
  • Pattern observations
  • Methodological insights
  • Future research directions

Tool Preferences

Visualization styles, output formats, and tool configurations

Auto-loaded

Storage:

.research/preferences.yaml

Examples:

  • Preferred visualization style
  • Output format (APA, Chicago, etc.)
  • Language preferences (EN/KO)
  • Agent behavior settings

Project Structure

Memory is organized in a .research/ directory within your project:

.research/ Directory Structure
.research/
├── baselines/
│   ├── baseline-001.yaml
│   └── baseline-002.yaml
├── changes/
│   ├── change-001.yaml
│   └── change-002.yaml
├── sessions/
│   ├── session-20250205-143022.yaml
│   └── session-20250205-153145.yaml
├── notes/
│   ├── literature-notes.md
│   └── methodological-insights.md
├── project-state.yaml
├── decision-log.yaml
├── preferences.yaml
└── checkpoints.yaml

All memory files are version-controlled and can be committed to git for team collaboration.

Context Keywords (Auto-Trigger)

When you use these keywords, relevant memory is automatically loaded:

Project Reference

Keywords:

my researchthis studyour project내 연구이 연구연구 진행

Loads:

Project ContextDecision Log

Session Continuity

Keywords:

continueresumewhere were we계속이어서어디까지

Loads:

Session MemoryActive Tasks

Decision Reference

Keywords:

we decidedI chosepreviously selected결정했던선택했던

Loads:

Decision LogCheckpoint History

Insight Recall

Keywords:

findingsinsightsobservations발견인사이트관찰

Loads:

Research Notes

Why Memory Matters

Context-aware memory provides critical advantages for research workflows:

No Repetition

Never re-explain your research context—it's always available to agents

Consistent Decisions

Past decisions inform future choices, ensuring methodological consistency

Audit Trail

Complete decision history for your methodology section

Team Collaboration

Version-controlled memory enables team members to stay synchronized

Session Resilience

Resume work seamlessly after interruptions or context window limits

Memory Commands

Explicit CLI commands to interact with memory:

/diverga:memory context

View current loaded memory state

/diverga:memory refresh

Reload memory from files

/diverga:memory clear

Clear session memory (keeps project memory)

/diverga:memory export

Export memory as markdown report

Ready to Experience Context-Aware Memory?

Start your research with memory that persists across sessions and agents.