SAIHM — Sovereign AI Horizontal Memory

What is SAIHM — Sovereign AI Horizontal Memory?

SAIHM is an MCP server that provides a sovereign, encrypted, sharable, persistent memory layer for AI agents. It exposes eight tools for storing, recalling, forgetting, sharing, and governing memory cells, with cryptographic auditing on COTI V2 mainnet. It is intended for users of MCP‑capable AI agents like Claude Code, Claude Desktop, and custom agents who want to own their memory.

How to use SAIHM — Sovereign AI Horizontal Memory?

Install the package via npm install @saihm/mcp-server or run directly with npx @saihm/mcp-server. Configure two environment variables (SAIHM_ENDPOINT_URL and SAIHM_AUTH_HEADER) pointing to a SAIHM operator endpoint, then wire the server into your MCP client (e.g., Claude Code) using the provided JSON configuration block.

Key features of SAIHM — Sovereign AI Horizontal Memory

• Eight MCP tools: remember, recall, forget, status, share, revoke share, governance propose, governance vote.
• True cryptographic erasure compliant with GDPR Article 17.
• Selective memory sharing with another agent or user.
• All operations forwarded to a SAIHM operator endpoint audited on COTI V2 mainnet.
• Reporting engine with bespoke templates for compliance (GDPR, SOC2, ISO27001, etc.).
• Security defaults: HTTPS-only endpoints, 30s abort window, 16 MB response cap, no filesystem reads.

Use cases of SAIHM — Sovereign AI Horizontal Memory

• Give any MCP‑capable AI agent a persistent, encrypted memory it can recall across sessions.
• Reduce context token usage by retrieving only relevant memory cells (benchmarked 62.8%–85.9% fewer input tokens).
• Comply with data erasure regulations by cryptographically forgetting specific memories.
• Share selected memories with another agent or user through