Sync

Moneypenny uses CRDTs (Conflict-free Replicated Data Types) at the SQLite level to sync data between agents. No central server required — agents merge state directly from each other’s database files.

Moneypenny exposes sync operations through MCP, so you can check status, push, pull, and connect to cloud sync with natural language in any MCP-compatible client (Claude Desktop, Cursor, etc.) or through the CLI.

What Syncs

Table	Syncs	Rationale
facts	Yes	Shared knowledge across agents
fact_links	Yes	Graph relationships between facts
skills	Yes	Reusable procedures propagate fleet-wide
policies	Yes	Governance rules propagate fleet-wide
Messages	No	Conversations are agent-local
Scratch	No	Session working memory is ephemeral
External events	No	Raw ingest data stays with the importing agent

How It Works

Each agent database has a unique site ID and a monotonic write clock. When you sync, CRDT metadata is used to merge changes without conflicts:

1. Export pending changes as a binary payload
2. Transfer the payload (local file copy or network)
3. Apply the payload to the target database
4. CRDT merge resolves any concurrent writes

Checking Status

CLI

mp sync status

Natural Language

Ask your MCP-connected agent:

Show me the sync status

Shows the site ID, database version, and per-table sync status.

Syncing

Bidirectional sync with all configured peers:

CLI

mp sync now

Natural Language

Ask your MCP-connected agent:

Sync with all peers now

One-way push (this agent → peer):

CLI

mp sync push --to research
mp sync push --to /path/to/other.db

Natural Language

Ask your MCP-connected agent:

Push my changes to the research agent

One-way pull (peer → this agent):

CLI

mp sync pull --from research
mp sync pull --from /path/to/other.db

Natural Language

Ask your MCP-connected agent:

Pull changes from research

Configuration

Configure peers and cloud sync in moneypenny.toml:

[sync]
tables = ["facts", "fact_links", "skills", "policies"]
peers = ["research", "ops-bot"]
interval_secs = 300   # auto-sync every 5 minutes; 0 = manual only
# cloud_url = "sqlite-cloud-url?apikey=..."

Cloud Sync

For remote sync without direct file access:

CLI

mp sync connect "your-cloud-sync-url?apikey=..."
mp sync now

Natural Language

Ask your MCP-connected agent:

Connect to cloud sync at your-cloud-sync-url?apikey=…

Cloud sync uses the same CRDT merge — the cloud backend is a relay, not an authority.

Scoped Knowledge

Sync moves rows, but visibility is controlled by fact scope:

• Private facts sync but remain invisible to agents below the trust threshold
• Shared facts are visible to all agents after sync
• Protected facts require elevated trust

Scope enforcement happens at query time via SQL filters — the sync layer moves all data; the policy layer controls who can read it.