Memory infrastructure for every customer you ship to
MemoryRouter gives each of your end users their own persistent memory vault — programmatically. One API key per customer. Isolated vaults. Usage-based billing across your entire user base.
Stop building multi-tenant memory from scratch. Plug in a battle-tested layer and ship.
The problem you keep hitting
If you ship AI agents to real users, you already know.
Your agents forget
Every session starts from zero. Users re-explain their context, their preferences, what happened yesterday. Retention dies on the "who are you again?" message.
Multi-tenant memory is hard
Per-user vector stores, isolation, key rotation, usage metering, billing — that's months of infra work before you write your first retrieval query.
Scaling breaks costs
Naive memory systems stuff everything into context. Token spend balloons. Margins collapse the moment a power user shows up.
How it works for enterprise
One integration. One API. Every customer gets their own isolated vault.
Your customer signs up in your app
Standard signup flow. No MemoryRouter account required on their side.
You mint them a vault + Memory Key
Provision a scoped mk_... key per customer from your enterprise dashboard. Each key is bound to an isolated vault for that user.
Their agent calls the API with their key
Point your OpenAI-compatible SDK at api.memoryrouter.ai and pass the customer's Memory Key. Persistent memory, automatic. No cross-tenant leakage. Ever.
You pay for actual usage
Metered across all your customers. One bill. Usage-based so you capture value at scale without flat-fee blowups.
curl https://api.memoryrouter.ai/v1/chat/completions \ -H "Authorization: Bearer mk_customer_9f3k..." \ -H "Content-Type: application/json" \ -d '{ "model": "openai/gpt-4o", "messages": [...] }'
- → mk_customer_9f3k identifies the customer's vault
- → Relevant memory is retrieved and injected into context
- → Request proxies to the provider (OpenAI, Anthropic, Google…)
- → Response is stored back in that customer's vault. Next call continues the thread.
Mint, rotate, and revoke keys per customer from your enterprise dashboard, or via the openclaw mr CLI for agent-based deployments.
Why enterprises choose us
Built for teams shipping memory to real customers — not a demo on a laptop.
Per-customer vaults
Strict data isolation between end users. SOC 2-aligned architecture. No shared indexes, no accidental cross-tenant retrieval.
Programmatic provisioning
Mint, rotate, and revoke API keys per customer via API. Plug it into your signup flow and forget about it.
Usage-based pricing
You pay for what your customers actually use — tokens in, tokens out. No flat-fee traps as you scale.
Committed-use discounts
Volume commits unlock significantly lower per-token rates. Predictable pricing as you grow.
Dedicated support
Shared Slack channel with the team that built it. Real engineers, not a ticket queue.
Custom SLA + contract
Uptime SLAs, DPAs, custom terms, MSAs — we sign paper. Enterprise procurement-friendly.
Enterprise pricing
Committed-use tiers. Every tier includes per-customer vaults and programmatic key provisioning.
Starter
$2,000/mo
committed
- • Up to 500M tokens / mo
- • Unlimited per-customer vaults
- • Programmatic API key provisioning
- • Email support
- • Standard SLA
Growth
$10,000/mo
committed
- • Up to 3B tokens / mo
- • Everything in Starter
- • Shared Slack support channel
- • 99.9% uptime SLA
- • Custom integration help
- • Committed-use rate discounts
Enterprise
Custom
let's talk
- • Unlimited tokens
- • Everything in Growth
- • Dedicated account manager
- • Custom SLA + DPA
- • SOC 2 report on request
- • On-prem / VPC deployment option
- • Custom data residency
Running smaller or just testing? Self-serve pay-as-you-go pricing →
Who this is for
If any of these sound like you, we should talk.
AI app companies
Shipping agents to thousands of end users. Every one of them needs their own memory, their own context, their own history.
Agencies
Building custom agents for clients. Each client gets isolated memory, you bill through one account, everyone's data stays separate.
Infra teams
You need memory as a service so your platform team can stop building vector stores and go back to shipping product features.
Built for buyers who ask hard questions
We know the procurement checklist. Here's where we stand.
Data isolation per vault — no shared tenancy at retrieval time
SOC 2 Type II in progress (Q3 2026 target)
DPAs available on request
Custom data residency on the Enterprise tier
Encryption at rest and in transit
Scoped API keys with rotation + revocation
Ready to add persistent memory to your product?
Quick intro call. We'll scope your volume, walk through the API, and send pricing the same day.