Configuration
Configure Diverga for your research workflow with configuration files, environment variables, and memory systems.
Overview
Diverga uses a multi-layered configuration system to maintain context across sessions and projects. Understanding these files helps you customize behavior and persist research decisions.
Configuration Files
Four core files control Diverga's behavior and memory:
CLAUDE.md
Project root
Project-specific instructions for Claude Code
Agent triggers, workflow patterns, domain knowledge
AGENTS.md
Project root (auto-generated)
Hierarchical codebase documentation
File summaries, architecture notes, component relationships
project-memory.json
.omc/project-memory.json
Persistent project context
Tech stack, build commands, conventions, directives
notepad.md
.omc/notepad.md
Session memory and working notes
Priority context, working memory, manual notes
Environment Variables
API keys and configuration for external services:
ANTHROPIC_API_KEYREQUIREDClaude API access (required for all agent operations)
sk-ant-api03-...GROQ_API_KEYOPTIONALGroq LLM for I-category screening (budget-friendly alternative)
gsk_...OPENAI_API_KEYOPTIONALOpenAI API for embeddings and optional agents
sk-proj-...CLAUDE.md Structure
The CLAUDE.md file defines project-specific behavior and agent triggers:
# CLAUDE.md
## Project Overview
This project uses the I-category pipeline for systematic literature reviews
in education research.
## Agent Triggers
- "systematic review" → I0-ScholarAgentOrchestrator
- "meta-analysis" → C4-MetaAnalyst
- "literature search" → B1-LiteratureScout
## Research Context
- Domain: Educational Technology
- Methodology: PRISMA 2020
- Analysis: Meta-analytic structural equation modelingProject Memory Schema
The project-memory.json file stores structured project context:
{
"techStack": {
"languages": ["Python", "TypeScript"],
"frameworks": ["Next.js", "FastAPI"],
"databases": ["PostgreSQL", "ChromaDB"]
},
"build": {
"install": "pnpm install",
"dev": "pnpm dev",
"build": "pnpm build",
"test": "pytest tests/"
},
"conventions": {
"codeStyle": "Black formatter, 100 char line length",
"naming": "snake_case for Python, camelCase for TypeScript",
"imports": "Absolute imports from src/"
},
"directives": [
{
"directive": "Always use TypeScript strict mode",
"priority": "high",
"context": "Type safety is critical for research data"
}
]
}Notepad Structure
The notepad.md file maintains session memory with three sections:
# Priority Context (always loaded, max 500 chars)
Research question: How do AI chatbots improve language learning?
Stage: PRISMA screening (Stage 3)
Database: Semantic Scholar + OpenAlex (N=342 papers)
# Working Memory (auto-pruned after 7 days)
[2025-02-05T10:30:00Z] Excluded 23 papers for non-English
[2025-02-05T11:15:00Z] Identified 5 high-impact papers (>100 citations)
[2025-02-05T14:20:00Z] Changed inclusion criteria: added "speaking skills"
# MANUAL (never auto-pruned)
Key finding: Chatbot effectiveness depends on feedback immediacy
Important papers: Smith (2023), Lee (2024) - both use similar frameworksSetup & Installation
Configure Diverga after installation:
1. Run setup wizard
/diverga:setupInteractive configuration for checkpoints, HUD, and language
2. Set API keys
export ANTHROPIC_API_KEY=sk-ant-...Add to .env file or shell profile
3. Initialize project
/diverga:deepinitGenerate AGENTS.md and initialize project memory
Best Practices
Recommendations for effective configuration:
- Keep CLAUDE.md concise - focus on project-specific patterns
- Use Priority Context for critical session state (research stage, current task)
- Add directives for non-obvious conventions (e.g., "Never use pandas.append()")
- Review notepad.md regularly and promote insights to project-memory.json
- Version control CLAUDE.md and AGENTS.md, but .gitignore .omc/