Your AI agent, with a memory that is yours.

SAIHM — Sovereign AI Horizontal Memory. A sovereign, encrypted, sharable, persistent memory protocol for AI agents.

The memory layer for businesses and regulated enterprises — and the developers shipping to them. Retail tools remember; SAIHM can prove what it remembers, shares, and erases.

Start free →   Join SAIHM →   Run a demo →

For you

Tell your agent “Join SAIHM.” Pay in everyday stablecoins. No technical setup.

For developers

One protocol your agent already speaks. Open source. Earn-via-build incentives.

For organisations

Regulatory-ready: portable, auditable, with right-to-erasure on demand.

For government

Sovereign data trail. Public-protocol auditability. No vendor lock-in.

What SAIHM is for

Today’s AI agents forget. Or worse, they remember on someone else’s server, under someone else’s rules. SAIHM gives any AI agent — commercial or open-source — a memory layer the user actually owns. Encrypted. Portable. Sharable when you choose. Erasable when you don’t.

  • Sovereign. The user holds the keys, not the agent vendor.
  • Encrypted. Every memory is sealed before it leaves the device.
  • Sharable. Grant access to a teammate, an auditor, or another agent — revoke it whenever.
  • Persistent. Memory survives the conversation, the app upgrade, the vendor switch.

For any AI agent, anywhere

SAIHM is not an AI agent. It is the memory layer that any AI agent can talk to — commercial assistants, open-source agents, or anything you build yourself. If your agent speaks the standard agent protocol, it speaks SAIHM.

See it run

Runnable demos ground a memory you own in every major model — Claude, GPT, DeepSeek, Qwen, Kimi, GLM — then prove you can erase it. Each runs offline in about a minute; no account needed to try it. There are drop-in adapters for LangChain and LlamaIndex, and an MCP server for Claude Code and Cursor.

Browse the demos →   developer guide →

Measured: ~80% fewer context tokens

Most AI agents re-send their entire transcript every turn, so the context you pay for grows quadratically as a session runs — and eventually overflows the context window. SAIHM recalls a small, bounded set of memory cells instead. Across a realistic multi-session coding task, that cut input tokens by 62.8% to 85.9%, and the longer the session, the wider the gap.

Session lengthRe-send everythingSAIHM recallFewer tokens
5 turns1,62860562.8%
10 turns6,0911,27379.1%
15 turns13,1752,02384.6%
18 turns18,6882,63285.9%

Input/context tokens only, summed across every turn (output tokens are identical under both strategies). Counted with the GPT-4 BPE tokenizer; runs fully offline and deterministic, so anyone reproduces the same result. It measures resend-vs-recall token volume, not any one provider’s bill.

Run the benchmark yourself →   or browse the demos →

Independent · on a public network

Get started

Pick the audience page that fits you. Each one tells you, in plain language, what to do next.

Start free →   Join SAIHM →