MemVault

Obsidian-powered persistent memory system for AI coding agents.

MemVault gives your project a living, queryable memory that survives across sessions, agents, and tools. Unlike plain markdown wiki systems, MemVault leverages Obsidian vault structure to prevent context overflow, enable instant search, and maintain cross-link relationships between decisions, patterns, and bugs.

Why MemVault?

| Problem | Plain Wiki | MemVault | |---|---|---| | Context overflow | All files read every session | Only relevant folder read | | Stale data | No timestamps | YAML frontmatter + last_updated | | Manual log folding | /memwiki-fold required | Daily log files auto-segment | | Search | Read all files | Full-text + Dataview queries | | Cross-references | Manual [[links]] | Obsidian backlinks + graph | | Orphaned content | No detection | /mv-lint finds broken links |

Quick Start

npx memvault init

Creates:

• .memory/vault/ — structured wiki with folders for decisions, features, logs, patterns, bugs
• .memory/vault/agents/ — 200+ specialized agency-agents across 15 divisions
• .obsidian/ — Obsidian config (daily notes, file explorer, core features)
• AGENTS.md — protocol instructions for AI agents
• kilo.json — Kilo Code config linking instruction files
• .kilo/command/ — slash commands for the vault

Folder Structure

[Project Root]
├── .obsidian/                    ← Obsidian vault config
│   ├── app.json
│   ├── core-features.json
│   └── plugins/
├── .memory/
│   ├── vault/
│   │   ├── hot.md                ← Immediate context (read first)
│   │   ├── index.md              ← Wiki table of contents
│   │   ├── decisions/            ← Architecture Decision Records (one per file)
│   │   │   └── index.md          ← Decisions TOC
│   │   ├── features/             ← Feature-specific wiki pages
│   │   │   └── index.md
│   │   ├── logs/                 ← Daily work logs (auto-segmented)
│   │   │   └── 2026-07-01.md
│   │   ├── patterns.md           ← Coding conventions
│   │   └── bugs.md               ← Known issues
│   │   └── agents/               ← 200+ specialized agency-agents (15 divisions)
│   │       ├── index.md          ← Master registry
│   │       ├── engineering.md    ← 30+ engineering agents
│   │       ├── design.md         ← 9 design agents
│   │       ├── finance.md        ← 5 finance agents
│   │       ├── security.md       ← 10 security agents
│   │       ├── product.md        ← 4 product agents
│   │       ├── testing.md        ← 8 testing agents
│   │       ├── marketing.md      ← 30+ marketing agents
│   │       ├── sales.md          ← 9 sales agents
│   │       ├── project-management.md  ← 7 PM agents
│   │       ├── specialized.md    ← 30+ specialized agents
│   │       ├── support.md        ← 6 support agents
│   │       ├── paid-media.md     ← 7 paid media agents
│   │       ├── spatial-computing.md  ← 6 spatial computing agents
│   │       ├── game-development.md   ← 20+ game dev agents
│   │       ├── academic.md       ← 5 academic agents
│   │       └── gis.md            ← 7 GIS agents
│   └── .raw/                     ← Immutable source docs (API docs, PDFs)
├── AGENTS.md                     ← Master protocol file
├── kilo.json                     ← Kilo Code config
├── .kilo/
│   └── command/
│       ├── mv-init.md            ← Initialize vault
│       ├── mv-ingest.md          ← Scan repo → populate wiki
│       ├── mv-lint.md            ← Health check (broken links, stale files)
│       ├── mv-fold.md            ← Archive old daily logs
│       ├── roast.md              ← Business idea stress-test
│       ├── council.md            ← Technical build audit
│       └── team.md               ← Auto-select agency-agents
└── .gitignore

YAML Frontmatter Convention

Every wiki file uses structured metadata:

---
title: "Use Supabase for Auth"
created: 2026-07-01
updated: 2026-07-01
tags: [architecture-decision, auth, supabase]
status: approved
related:
  - [[bugs/known-issue-supabase-rate-limit]]
---

Agents parse frontmatter to understand context without reading the full file.

Obsidian Integration

MemVault is designed to work as an Obsidian vault. Install these plugins for full functionality:

• Core Plugin: Daily Notes — auto-creates logs/YYYY-MM-DD.md
• Core Plugin: Dataview — query wiki data like a database
• Community Plugin: Templater — auto-fill frontmatter on new files

Dataview Queries

Show recent decisions:

TABLE status, updated AS "Last Updated"
FROM ".memory/vault/decisions"
WHERE status = "approved"
SORT updated DESC
LIMIT 10

Find stale hot cache:

TABLE last_updated AS "Updated"
FROM ".memory/vault/hot.md"
WHERE last_updated < date(today) - 7d

Slash Commands

| Command | Description | |---|---| | /mv-init | Initialize MemVault in a project | | /mv-ingest | Scan repo → populate wiki folders | | /mv-lint | Health check — broken links, stale files, orphaned content | | /mv-fold | Archive old daily logs (keep last 30 days) | | /roast | Business idea stress-test (5 personas) | | /council | Technical build audit (7 personas) | | /team | Auto-select and spawn relevant agency-agents (200+ experts) |

Agency Agents Integration

MemVault integrates with agency-agents — a collection of 200+ specialized AI agent personalities across 15 divisions:

• Engineering (30+): Backend Architect, Frontend Developer, DevOps Automator, AI Engineer, Database Optimizer, Code Reviewer, Software Architect, SRE, Data Engineer
• Design (9): UI Designer, UX Researcher, UX Architect, Brand Guardian, Visual Storyteller
• Finance (5): Financial Analyst, Investment Researcher, FP&A Analyst, Bookkeeper, Tax Strategist
• Security (10): Security Architect, Penetration Tester, Incident Responder, Threat Intelligence Analyst
• Product (4): Product Manager, Sprint Prioritizer, Trend Researcher, Feedback Synthesizer
• Testing (8): Performance Benchmarker, API Tester, Accessibility Auditor, Test Results Analyzer
• Marketing (30+): Growth Hacker, SEO Specialist, Content Creator, Social Media Strategist
• Sales (9): Outbound Strategist, Deal Strategist, Sales Engineer, Pipeline Analyst
• Project Management (7): Studio Producer, Project Shepherd, Experiment Tracker
• Specialized (30+): Agents Orchestrator, Workflow Architect, MCP Builder, Automation Governance Architect
• Support (6): Support Responder, Analytics Reporter, Finance Tracker, Executive Summary Generator
• Paid Media (7): PPC Campaign Strategist, Search Query Analyst, Ad Creative Strategist
• Spatial Computing (6): XR Interface Architect, visionOS Spatial Engineer
• Game Development (20+): Game Designer, Unity Architect, Unreal Systems Engineer
• Academic (5): Anthropologist, Historian, Psychologist, Narratologist, Geographer
• GIS (7): GIS Analyst, Spatial Data Engineer, GeoAI/ML Engineer

Use /team to auto-select relevant experts based on your project type, or read .memory/vault/agents/ for the complete registry.

How It Works

1. Session Start: Agent reads hot.md (immediate context) → index.md (TOC) → relevant folder only
2. During Work: Agent updates logs/YYYY-MM-DD.md (daily log) + relevant decision/feature file
3. Session End: Agent updates hot.md with next steps + appends to today's daily log
4. Obsidian: Graph view shows connections. Dataview queries surface relevant context.
5. Agency Agents: Use /team to spawn specialized experts from the 200+ roster for deep domain expertise.

License

ISC