Why Local-First?

Memvault is built on local-first principles: your data lives on your machine, under your control.

The Philosophy

You Own Your Data

AI conversations are intellectual assets:

• Capture your reasoning process
• Document decisions and context
• Contain sensitive information
• Represent hours of work

They should belong to you, not the platform.

Problems with Cloud-First

1. Vendor Lock-In

• Data trapped in proprietary formats
• Can't export full history
• Platform changes affect access
• Risk of service shutdown

2. Privacy Concerns

• Conversations analyzed for training
• No guarantee of deletion
• Third-party access possible
• Data residency unclear

3. Limited Control

• Can't customize storage
• Search depends on platform
• No offline access
• Rate limits and quotas

4. Portability

• Moving between agents is hard
• Historical context doesn't transfer
• Must rebuild context each time
• Walled gardens

Local-First Benefits

1. Complete Ownership

Your Machine
├── All conversations
├── Full search index
├── Agent memories
└── Export formats

• No platform can delete your data
• No rate limits
• No usage analytics
• No third-party access

2. Privacy by Default

• Data never leaves your machine
• No cloud processing
• No AI training on your conversations
• You control who sees what

3. Offline-First

• Works without internet
• Search is instant
• No network latency
• No cloud downtime

4. Flexible Sync

You choose how to sync:

# iCloud
ln -s ~/Library/Mobile\ Documents/... ~/memvault

# Dropbox
ln -s ~/Dropbox/memvault ~/memvault

# Git
cd ~/memvault && git init

# Syncthing
syncthing add ~/memvault

Memvault never implements sync—you use the tool you already trust.

Architecture Implications

No Cloud Dependency

┌─────────────────────────────────────┐
│         Your Computer               │
│                                     │
│  Agent ←→ Memvault ←→ SQLite       │
│              ↓                      │
│         Markdown Files              │
└─────────────────────────────────────┘

• Agent talks to local process
• Data stored locally
• Search index local
• Sync optional

Open Formats

Archives: JSON + Markdown

• Human-readable
• Version control friendly
• Easy to export
• Future-proof

Memories: Plain Markdown

• Edit in any text editor
• Version with git
• Search with grep
• Platform-agnostic

Portable

Move between:

• Machines (sync files)
• Agents (open formats)
• Operating systems (cross-platform)
• Tools (grep, jq, cat work)

Real-World Scenarios

Scenario 1: Company Policy

"We can't use cloud AI that processes our internal data."

✅ Solution: Local archive + company NAS sync
All processing local, sync via corporate infrastructure.

Scenario 2: Sensitive Work

"My conversations contain trade secrets and customer data."

✅ Solution: Never leaves your machine
No risk of data leakage, full audit trail.

Scenario 3: Long-Term Value

"What if Claude Code shuts down in 5 years?"

✅ Solution: Archives in open formats
Data remains accessible, searchable, portable.

Scenario 4: Research

"I want to analyze my AI interaction patterns."

✅ Solution: Direct file access
Use any tool: Python, SQL, jq, Unix utilities.

Sync Strategy

You Decide

Memvault provides the local store. You choose sync:

Tool	Use Case	Setup
iCloud	macOS/iOS	Link to iCloud folder
Dropbox	Cross-platform	Link to Dropbox folder
Git	Version control	git init in vault
Syncthing	P2P sync	Add vault to Syncthing
rsync	Manual backup	Script to NAS
None	Single machine	No setup

Conflict Resolution

Because sync is external, conflicts are handled by your sync tool:

<!-- iCloud/Dropbox conflict markers -->
<<<<<<< Local
User prefers Go
=======
User prefers Rust
>>>>>>> Remote

Memvault detects conflicts and flags them for review.

Comparison

Cloud-First (ChatGPT, Claude.ai)

❌ Data on vendor servers
❌ Proprietary export
❌ No offline search
❌ Platform lock-in
❌ Privacy policy changes

Local-First (Memvault)

✅ Data on your machine
✅ Open formats
✅ Offline-first
✅ Platform-agnostic
✅ You set the rules

Trade-Offs

What You Give Up

• No built-in sync service
• Must manage backups yourself
• Cross-device requires setup
• More disk space used

What You Gain

• Complete control
• Total privacy
• Unlimited storage
• Custom workflows
• Data portability

Philosophy in Practice

Design Decisions

1. SQLite not cloud DB → offline, fast, portable
2. Markdown not proprietary → human-readable, greppable
3. Files not API → direct access, version control
4. Hooks not webhooks → local process, no network
5. MCP not cloud service → local socket, instant

Future-Proofing

Memvault's formats will work decades from now:

# In 2046, these still work:
cat memories/background/*.md
jq '.role' sessions/*/messages/*.json
grep -r "search term" sessions/

Getting Started

See Installation to begin using Memvault locally.

Next Steps

• Data Sovereignty
• Storage Design
• Multi-Device Sync