How it works

A memory layer that knows where to live, where to listen, and where to speak.

The architecture in five layers, plain English.

Storage

Your memories live on dedicated infrastructure. Encrypted at rest, encrypted in transit. Backups every 24 hours, retained for seven days. SOC 2 Type 1 and HIPAA compliant. Bring-your-own-key encryption available on the Family-Office tier.

The memory engine

An enterprise-grade memory engine, configured per client. Handles extraction, deduplication, retrieval, relevance scoring. We selected it for security posture and reliability. We manage it end-to-end. You never touch it.

The MCP server (cross-platform interface)

A Model Context Protocol server we configure to bridge your memory to every AI you use. Standardized protocol means every new AI that ships will connect to your memory without additional work on your side.

Your AI surfaces

Claude (web, desktop, Claude Code, Cowork), ChatGPT (web, ChatGPT Connectors as they ship), Gemini (Gemini Connectors as they ship), Grok (manual fallback), and every MCP-compatible assistant that ships next.

The concierge layer

The reason this works long-term. We do not hand you a tool and walk away. Avihay personally onboards you, tunes your memory quarterly, and is on call for anything that matters.

Account provisioning on dedicated infrastructure. Your private instance is spun up before the call. We do not start from zero on day one.
Encryption key setup (BYOK if Family-Office). We generate and store keys with you, in a process you control end-to-end.
MCP server configuration for your AI surfaces. Claude, ChatGPT, Gemini, anything MCP-compatible. We do every install.
Initial memory seeding (50-100 memories from your real work). We pull from emails you share, calendar context, prior AI conversations.
Test queries across each connected AI. We do not leave the call until your memory works on every surface.
Dashboard access for you (read-only by default). See what is in your memory at any time. Read-only protects the integrity of the layer.
Direct channel setup. WhatsApp or Telegram for Family-Office, email for Concierge. We tell you when we are on call.
Handoff document. A one-page reference for what you have, what you can do, who to contact.

AI surface	Today	Coming
Claude (web, desktop, Code, Cowork)	Full MCP integration	(current)
ChatGPT (web)	Connectors as they ship in your region	Q3 2026 estimate
ChatGPT (API / custom)	Full integration via SDK	(current)
Gemini (web)	Connectors as they ship	Q3 2026 estimate
Grok	Manual fallback (system prompt export)	Direct integration when xAI ships MCP
Perplexity	Manual fallback	Pending
Custom AI builds (your in-house tools)	Full integration via SDK or REST	(current)

We do not provide 24/7 SLA support. Same-business-day IDT response, on-call for actual incidents. If you need 24/7 globally, we are not the right fit.

We do not provide custom AI development. We integrate with your existing AI tools. If you need a custom model built from scratch, that is a separate engagement.

We do not retain your data after termination. Full export within 7 days, deletion within 14. Your data was always yours.

20 applications open.

First cohort onboarding through July 2026. We review every application personally within 72 hours.

Apply now →