Memory for the agents you ship.

What you don’t have to build

• No infrastructure. The protocol provides the storage tier, the relayer, the audit trail, and the on-chain anchor. You write zero memory-backend code.
• No human capital. No on-call rotation for a memory service. No key-management staff. No compliance team ramping up GDPR Art. 17 erasure procedures.
• No ongoing maintenance. Apache 2.0 source. Build commitments are dated and anchored on a public chain — you can verify what is running at any time.

Why standardisation matters here

Memory is becoming the agent differentiator. Every vendor reinventing memory privately is rebuilding the same four hard parts — sovereignty, erasure, sharing, and audit — badly, in isolation. A shared memory protocol means:

• Agents can hand off context. Your assistant can pass a sharable, user-consented memory cell to another vendor’s assistant without exporting raw text.
• The user holds the keys. You never store user PII in your tenant just to give your agent a memory. That eliminates a class of breach liability you do not need.
• Adoption goes up. Lower friction means more end-users adopt agentic products. Standardisation lifts the whole industry — including your share.
• Lower token cost at scale. Recalling a bounded working set instead of re-sending each session’s full transcript is measured at ~80% fewer context tokens on long runs (open benchmark) — across a fleet of agents, that compounds.

How to integrate

Three paths, in increasing order of integration depth:

• Drop-in. Your agent calls saihm_remember / saihm_recall over the existing MCP transport. The user opts in; your code is unchanged elsewhere. Hours, not weeks.
• Pre-integrated. Ship your agent product with SAIHM enabled by default (user opt-in on first run). One protocol, one onboarding flow, every customer covered.
• White-labelled. Use SAIHM as the memory primitive under your brand. Your UI/UX, your tier configuration via the operator-policy overlay, your support surface. SAIHM remains the protocol identity; the experience is yours.

Real-world examples

Names invented; scenarios drawn from how the protocol actually fits AI-product builders.

• A 12-person productivity-AI startup. Building a personal-assistant product. Instead of shipping a memory backend in v1, they pre-integrate SAIHM and ship the assistant. Memory becomes a checkbox in onboarding rather than a service to operate. Their compliance work shrinks because they no longer custody user PII for memory purposes.
• An established voice-AI vendor. White-labels SAIHM under their brand. Users opt in once at first run; the vendor’s UX is unchanged; their support surface no longer covers a memory subsystem they used to build and maintain in-house.
• A vertical AI platform for legal firms. Their customers are firms that cannot accept opaque vendor memory of privileged communications. SAIHM is the memory layer because cryptographic erasure receipts and per-record sealing align with the bar-association expectations the firms operate under.
• An AI gateway / model-API reseller. Offers a “bring-your-own-memory” option to customers who reuse the gateway across multiple frontier models. SAIHM is the memory option that does not bind the customer to the gateway’s commercial fate.

Pricing

Public pricing — per-call PAYG and monthly subscription tiers — is published on /pricing. Settlement in fiat via Stripe (card and other supported methods, subscription tiers) or in USDC.e on COTI V2 mainnet. Public pricing applies to all users.

What this isn’t

• Not a managed agent platform. SAIHM is the memory layer only — you keep your agent, your model, your UX.
• Not a hosted SaaS with opaque internals. The protocol runs against the public COTI V2 chain; build commitments are reproducible from the on-chain anchor.
• Not a competitor to your model or your product. Memory is orthogonal to inference.

Try the protocol

SAIHM is an autonomous Apache-2.0 protocol — settle a tier on-chain without contacting anyone. Public pricing on /pricing. Governance participation accrues automatically via paid use; see /governance.

Try SAIHM →