Overview
The memory system is Drevon’s most powerful feature. It gives AI agents persistent context across sessions — they remember what happened yesterday, what decisions were made, and what patterns emerged.
Without memory, every AI session starts from zero. With memory, agents compound knowledge over time.
How It Works
Session Start
Agents are instructed to read all memory files before doing any work. This restores full context from previous sessions.
During Work
Agents use memory context to make informed decisions — avoiding past mistakes, following established patterns, staying consistent with previous choices.
Session End
Agents write back significant learnings, decisions, and observations to the appropriate memory files.
Self-Evolution
When agents notice they’re performing the same multi-step workflow repeatedly, they create reusable prompts in .drevon/prompts/ — automating their own improvement.
Memory Files by Mode
Hub mode uses 4 memory files designed for cross-project awareness:user.md
Stores your personal preferences, communication style, and confirmed decisions.# User Preferences & Profile
## Communication Style
- Prefers concise, direct responses
- Likes code examples over explanations
## Confirmed Preferences
- Always use TypeScript strict mode
- Prefer Tailwind over CSS modules
## Decisions Made
- 2026-02-28: Chose Next.js 15 for all web projects
projects.md
A registry of all projects in the workspace with their status.# Project Registry
## Active Projects
### landing-page
- **Folder:** workspace/landing-page/
- **Status:** active
- **Started:** 2026-02-28
- **Summary:** Marketing landing page. Next.js 15 + Tailwind.
systems.md
Documents infrastructure, systems, and technical conventions.# Systems & Infrastructure
## Deployment
- All web projects deployed via Vercel
- Domain: *.trysudosu.com
## Conventions
- ESM only, no CommonJS
- Zod for all runtime validation
log.md
Append-only chronological record of significant actions.# Action Log
### 2026-02-28 — Workspace initialized
**Action:** Initialized AI-native workspace with drevon
**Outcome:** Memory system, agent configs created
### 2026-02-28 — Landing page shipped
**Action:** Built and deployed marketing landing page
**Outcome:** Live at drevon.trysudosu.com
Project mode uses 4 memory files scoped to the codebase:context.md
Project overview, architecture, key files, and dependencies.# Project Context
## Overview
E-commerce API built with Express + PostgreSQL
## Architecture
- REST API with controller/service/repository pattern
- PostgreSQL with Prisma ORM
- Redis for caching
## Key Files
- src/server.ts — Express app entry
- prisma/schema.prisma — Database schema
decisions.md
ADR-style decision records with context and rationale.# Technical Decisions
### 001 — Use Prisma over TypeORM
- **Date:** 2026-01-15
- **Context:** Need an ORM with good TypeScript support
- **Decision:** Prisma for type-safe queries and migrations
- **Rationale:** Better DX, schema-first approach
patterns.md
Code patterns, conventions, and anti-patterns.# Code Patterns
## Naming Conventions
- Files: kebab-case
- Classes: PascalCase
- Functions: camelCase
## Error Handling
- Always use custom error classes
- Never catch and silently ignore errors
log.md
Same append-only format as hub mode. Write Rules
Agents follow specific rules when updating memory:
| File | Write Mode | Format |
|---|
log.md | Append only | ### YYYY-MM-DD — title (newest at bottom) |
user.md | Update in place | Edit the relevant section |
projects.md / systems.md | Update or add | Update existing entry or add new |
context.md / decisions.md / patterns.md | Update or add | Update existing entry or add new |
log.md is append-only — agents should never delete or modify existing entries. This ensures a reliable audit trail.
Custom Memory Files
You can add custom memory files beyond the defaults:
{
"memory": {
"enabled": true,
"directory": ".drevon/memory",
"files": ["user.md", "projects.md", "systems.md", "log.md"],
"customFiles": ["competitors.md", "roadmap.md"]
}
}
Disabling Memory
If you don’t want persistent memory:
{
"memory": {
"enabled": false
}
}
npx drevon init --no-memory
Disabling memory removes one of Drevon’s most powerful features. Consider keeping it enabled — even light memory usage compounds significantly over time.
