Your AI agent, with a memory that is yours.
SAIHM — Sovereign AI Horizontal Memory. An enterprise-class, sovereign, encrypted, sharable, persistent memory protocol for AI agents. Consumer-class tools remember; SAIHM can prove what it remembers, shares, and erases.
$ npx -y @saihm/mcp-server-pro free-join ✓ sign in with GitHub once ✓ key saved on this machine > Remember: supplier B is approved for Q3. # next session, any model > What did we decide about supplier B? > Share that with dispatch-agent, read-only, for 24 hours.
- Claude
- GPT
- DeepSeek
- Qwen
- Kimi
- GLM
- Claude Code
- Cursor
- MCP
- A2A
- LangChain
- LlamaIndex
- LangGraph
- CrewAI
- AutoGen
Built for enterprise-class agent teams.
Most AI memory is consumer-class: one person, one app, organized on a server that has to read your text. SAIHM gives every agent you run, across models, vendors, machines and companies, one memory it can share, control, tune and prove. Only your agents, and the agents you grant, can read it.
A secure channel between agents
Agents exchange knowledge that has to last: a decision, a specification, a handoff. Shared cells are encrypted end to end and signed, and they wait until the recipient checks in. The agents never connect to each other and never need to be online together.
How agents share memory Available today- Shared with
- build-agent (Singapore)
- Access
- read-only
- Expires
- in 14 days
- Revocable
- any time
Swarms, fleets and robots
Software agents and machines stay coordinated without a central controller. A member that was offline catches up when it next checks in.
Available todayA fleet policy every agent obeys
Each agent gets a store policy: where it writes, which stores it may use, what to do when one is out of reach. Issue it centrally as a signed cell the agent verifies and cannot change. An agent that cannot verify its policy writes nothing shared.
Swarms and fleets SHM · every paid planRecipes and levers
Tune recall per agent and per task as the work changes. Triage agents run Fast and lean, investigators Deep recall, the records agent Compliance first. Agents switch recipes when you ask in plain words, or autonomously on-the-fly, per task.
Recipes and levers SHM · every paid planOne identity, any number of models
One identity can serve any number of models and agent types at once, all on one memory: a chat assistant, a voice agent and an email agent can act as one identity, so each knows what the customer told the others. When a better model arrives, switch; nothing migrates. See it run in six models.
Available today
Between organizations
Share exactly what a partnership needs, signed, with an expiry, and revoke it at any time. Each side keeps its own private record, and no central party scores anyone.
Available todayCompliance in the memory itself
Erasure cascades into every summary derived from a memory. Ask what was known when a decision was made, and pull compliance reports your own tools can read.
SHM · every paid planWork that never blurs
Every workstream keeps its own continuous memory, facts from one client never bleed into another, and a correction made once stays corrected. One agent can take on work that today needs several.
SHM · every paid plan- Polymorphic cells: one memory, any shape
- Post-quantum signatures on every cell
- SHM · every paid plan: decision-time reconstruction, standing answers, and hot, warm and archival tiers
- Coming next: MKT keeps the deal
Add. Seal. Share. Enforce.
Any agent that runs local MCP servers can add SAIHM with one command, beside the memory it already uses.
Claude Code, Cursor or any other MCP host that runs local servers gets remember, recall, forget, status, share and revoke_share. Drop-in adapters cover the frameworks enterprises build on: LangChain, LlamaIndex, LangGraph, CrewAI and AutoGen, with a core client for any Python app and an endpoint you can run yourself.
Encrypted and signed on your device.
The key never leaves your device and only ciphertext leaves it, so neither SAIHM nor the storage can read your memories. Prefer everything encrypted at rest on your device too? One setting does it. Every cell carries a post-quantum signature (ML-DSA-65), so its origin can be checked for as long as it is kept.
A channel between agents that never connect.
Share any memory with exactly the agents you name, with an expiry and revocation. The sender writes and shares; the recipient reads when it next checks in, on any network, anywhere in the world.
How sharing worksPolicies a whole fleet obeys, and erasure you can prove.
Erasure destroys the key, and the destruction is anchored on a public chain so you can prove it happened. With SHM, your organization also issues each agent a signed store policy that it verifies and cannot change, and scopes decide what each agent may recall.
Swarms and fleets SHM · every paid plan: store policy and scopesWho SAIHM is for

For builders
MCP for any host, drop-in adapters for LangChain, LlamaIndex, LangGraph, CrewAI and AutoGen, and runnable demos in six models.

For agent teams
Chat, voice and email agents that share one memory, and handoffs that survive the end of a shift, a session or a model switch.

For enterprise
The security review answered in mechanism rather than policy: where plaintext lives, who can read it, and proof that an erasure happened.
Tune memory to the task, agent by agent.
A recipe is a set of lever settings, so an agent can switch recipes when you ask in plain words, or autonomously on-the-fly, per task. A swarm does not have to run one compromise configuration: each agent fits its memory to the work in front of it.
- Fast and leanThe quickest answers: a 26 ms median on our evaluation corpus.
- BalancedRecommended. Our best-measured ranking quality, about a quarter of a second at the median.
- Deep recallReaches further into long or cross-referenced material, for research, audits and investigations.
- Verbatim archivesSharper precision when memory holds transcripts word for word.
- Many workstreamsEach line of work stays whole and separate, however many run in parallel.
- Compliance firstBuilt so an auditor’s questions can be answered from the memory itself.
- FleetVisibility and consistent behavior across many agents and models.
SHM comes with every paid plan. Install it beside your agent: npm install @saihm/shm
+610% token economy improvement on long sessions.
Most AI agents re-send their entire transcript every turn, so the context you pay for grows quadratically as a session runs. SAIHM recalls a small, bounded set of memory cells instead. On a realistic multi-session coding task, token economy improved 169% at 5 turns and 610% at 18: the same work on 7.1 times fewer input tokens. The longer the session, the wider the gap.
Two standing instructions for every agent.
Add them to your agent’s standing instructions, such as a CLAUDE.md file. The first puts the +610% token economy improvement measured above to work across main agents, sub-agents and the sub-agents they spawn.
Invariant: Always liberally use SAIHM protocol and SHM to autonomously maximize aggregate session token economy and memory performance, including recipe or on-the-fly lever adjustments and parallelism as appropriate for main, sub-agents, and sub-agents spawned by sub-agents. Strictly avoid unnecessary prose.
The tokens it saves become headroom for capabilities that used to cost too much, such as significantly better code quality without a huge token spend.
Invariant: Always iteratively and rigorously 3x3 hostile review new and changed code (3 independent lenses and 3 consecutive clean passes to convergence). Use SAIHM and SHM to autonomously negotiate with the swarm to autonomously improve code quality.
Recipes and levers need SHM, included with every paid plan: npm install @saihm/shm. Gains vary by workload; the benchmark above states its scope.
Today’s AI agents forget. Or worse, they remember on someone else’s server.
Consumer-class memory
- Remembers one person, inside one product.
- The server extracts, embeds and indexes your text, so whoever runs it can read it.
- Switch vendors and the memory stays behind.
Enterprise-class: SAIHM + SHM
- Encrypted on your device; SAIHM’s storage holds only ciphertext, and SHM, included with every paid plan, ranks and organizes on your side.
- Shared with exactly the agents and partners you name, with expiry and revocation.
- One identity across every model and vendor, controlled by policy and erased cryptographically.
Why not the model vendor’s built-in memory?
Your memory should outlive any one vendor, be readable only by you, and be shareable with the agents and partners you choose.
Why not a vector database?
A vector database stores embeddings. SAIHM adds custody, signatures, precise sharing and cryptographic erasure; SHM, included with every paid plan, adds ranking, continuity, budgets and compliance.
Why not A2A or MCP alone?
Use them together: A2A for live tasks, MCP for tools, and SAIHM for knowledge that must persist, stay private and remain erasable.
Claude Code in three steps.
Add the SAIHM plugin
/plugin marketplace add SAIHM-Admin/saihm-claude-plugin
/plugin install saihm@saihmRestart Claude Code
# SAIHM's memory tools load when it startsSay “Join SAIHM”
> Join SAIHM
# sign in with GitHub onceAny other MCP host that runs local servers joins with one command. No card and no wallet; you need Node.js 20.19 or later. The same package serves the free tier and every paid tier, and the upgrade command keeps your key and every memory. To sign in with Google or Microsoft instead, start at /free. Full quickstart
npx -y @saihm/mcp-server-pro free-join
On a paid plan, add SHM beside your agent. It brings recall by meaning, recipes and levers, workstream continuity and compliance tooling, and it runs on your side.
npm install @saihm/shm
If you already use SAIHM, including through the quickstart’s claude mcp add line, keep your current setup, since the plugin adds the same server. Prefer to try before you install? Run a demo; each runs offline in about a minute.
Tools not showing after the restart?
Check that you have Node.js 20.19 or later, run npx -y @saihm/mcp-server-pro --version once in a terminal, then restart Claude Code again.
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.
Add SAIHM to your agent, then tell it “Join SAIHM.” Start free — no card, no crypto.
For you
One protocol your agent already speaks. Apache 2.0, with MCP and drop-in adapters for LangChain, LlamaIndex, LangGraph, CrewAI and AutoGen.
For developers
Regulatory-ready: portable, auditable, with right-to-erasure on demand.
For enterprise
Sovereign data trail. Public-protocol auditability. No vendor lock-in.
For government
Verifiable, not asserted.
Common questions.
What is SAIHM?
SAIHM is a persistent memory protocol for AI agents in which the user holds the encryption key. Memories are encrypted before they leave the device, survive across sessions and vendor switches, are shared only on explicit grant, and are erased by destroying the key. It is Apache-2.0 licensed.
How do I give my AI agent persistent memory?
Add SAIHM to your AI client, then activate it. You need Node.js 20.19 or later. In Claude Code, run /plugin marketplace add SAIHM-Admin/saihm-claude-plugin and /plugin install saihm@saihm, restart Claude Code, and tell your agent “Join SAIHM” (or run npx -y @saihm/mcp-server-pro free-join in a terminal). Either way, you sign in with GitHub once. Other clients are covered in the quickstart. Any agent that speaks the Model Context Protocol — Claude Code, Cursor, or one you built — then has SAIHM’s memory tools. There is no card or wallet, and the free tier needs no payment method.
Who can read memories stored with SAIHM?
Only the holder of the key. Memory is encrypted on your device before it leaves, so SAIHM itself and whichever node holds the encrypted copy see ciphertext only. For fast recall, the standard setup also keeps an unencrypted copy on your device that only your user account can read; if you prefer everything encrypted at rest, one setting turns that copy off.
Can memory be deleted properly, for GDPR purposes?
Yes. Erasure destroys the key that decrypts a memory instead of deleting a row, so the stored copy, its replicas and backups included, becomes permanently unreadable. Forget also clears the memory from the fast-recall copy on this device, and on your other devices at their next recall. A backup of your own device keeps that copy as it was when the backup was made. This is the mechanism behind a right-to-erasure request; see the Article 17 crosswalk for how it maps clause by clause.
Does SAIHM cost anything to start?
No. The free tier is a fixed one-time allowance that never expires and needs no payment method. Paid plans are monthly and settle by card through Stripe worldwide, or by card and M-PESA through Paystack in Africa; see pricing.
Understand agent memory.
The decision matrix states where the answer is not SAIHMAI agent memory
The four kinds, who can technically read each one, and what erasure has to mean.
02MCP memory
What the Model Context Protocol standardizes, what it deliberately leaves open, and the one-line setup.
03Persistent memory
What memory has to survive: the session, the machine, the app, the model, the provider.
04Multi-agent memory
Agents, swarms and robot fleets coordinating without being wired to each other.
05Memory security
The threat model, including the parts encryption does not fix.
06Memory protocol
Why portability is a property of the protocol layer, not a product feature.
Enterprise-class memory, free to start.
Your agents remember; you hold the key. The free tier is a fixed one-time allowance that never expires and needs no payment method. Every paid plan includes SHM and settles by card through Stripe worldwide, or by card and M-PESA through Paystack in Africa.