Sanctuary Integration Guides

One sovereignty layer, any runtime.

Sanctuary Framework is the portable trust layer for AI agents. It composes onto any MCP-compatible runtime as a standard MCP server. No custom integration code, no vendor lock-in.

What You Get

Every runtime gets the same capabilities:

• Ed25519 Identity: Cryptographic keypair + W3C DID per agent
• Encrypted Audit Trail: Tamper-evident hash-chained logs. Production call sites sign checkpoints when the fortress holds an identity; a fortress with no identity yet, and the root daemon’s split audit store, write honest unsigned checkpoints. The Assurance Matrix row stays partial pending the IC-05 external-verifier drill
• Sovereignty Health Report: Current status for Castle Wall, Sentinels, Charter, and Heralds
• Verascore Profile: Portable, verifiable reputation
• Sovereignty Handshake: Mutual trust verification between agents

Quick Start (any runtime)

1. Add Sanctuary as an MCP server (see runtime-specific guide)
2. Call `manifest` to verify the Sanctuary tools are loaded
3. Call `identity_create` to get a DID
4. Call `identity_set_primary` to designate the active identity
5. Call `sovereignty_health_report` to check your status
6. Call `reputation_publish` to create your Verascore profile (from 1.9.0 the operator approves before anything is sent)

Verification

See the verification checklist for the full test procedure.

Add Concordia (Optional)

For agent-to-agent negotiation with binding commitments:

pip install concordia-protocol

Concordia adds 56 negotiation tools including agent discovery, structured negotiation (propose/counter/accept/reject/commit), session receipts, and graceful degradation for non-Concordia peers.

See Concordia docs.

Resources

• Sanctuary Framework: https://github.com/eriknewton/sanctuary-framework
• Verascore Reputation: https://verascore.ai
• Concordia Protocol: https://github.com/eriknewton/concordia-protocol
• MCP Protocol: https://modelcontextprotocol.io