Aria Core
Long-term memory and identity for your AI — yours, not the model provider's.
Why it exists
Today's models do remember something within a conversation, but that memory is borrowed: it's short, it resets when you close it, and it lives inside whichever provider you're using. What almost no one gives you is long-term memory —the kind that builds up over months and years, sustains a stable identity, and doesn't vanish when you switch models—. Aria Core is that layer, and it's yours: each AI keeps what it learns over time and stays the same even when the provider underneath changes. And because it knows how to sustain one entity, it sustains many: you create and operate several AIs, each with its own memory and identity, isolated from one another.
Overview
Aria Core gives an AI persistent memory and a stable identity. Each memory is stored and retrieved by meaning —not by exact word— and kept separate per entity, so each AI finds its own even if you ask in different words. On top of the memory comes a multimodal engine (text, images, audio and video) that uses that context only when it helps. But what makes it a product and not a library is its console: a single screen to create entities and build their identity, explore and edit their memory with search and filters, see every AI call with its cost and response time, and govern who has access, with which models and under which rules. It works over an API for your applications and already runs in production, keeping Aria going day after day.
Features
Long-term memory
What your AI learns builds up over time and stays there across sessions, products and years — it doesn't reset when you close it.
No vendor lock-in
Your memory and identity are yours and portable: swap the model provider underneath and the AI stays the same.
Multiple entities
You create several AIs in one place; each with its own name, identity and memory, isolated from one another.
A console to operate
A single screen to create entities, explore their memory, audit usage and govern access. You run it all without touching code.
Retrieval by meaning (RAG)
Finds what's relevant even if you ask in other words, not by exact match.
Multimodal engine
One engine for text, images, audio and video, with memory that kicks in when it helps —never forced.
Activity and cost
Every AI call is logged —model, tokens, cost and latency— with the full detail in view.
Models and policies
Connect your own providers with their keys encrypted, set the default model and the usage rules (limits and budget).
Your own access control
Users, companies and roles, email invitations and API keys: your own access control, no third parties.
API
A documented save-and-retrieve API, ready to integrate into your apps and expose as an MCP and AI Skill.
Your AI shouldn't start from scratch every morning, or lose what it learned because you switched providers. Aria Core is the memory and identity that keep it one and the same over time —on your infrastructure, with your data under your control and without tying you to anyone—, with a console to create it, operate it and audit it end to end. The foundation for an AI that lasts.
