01
Recall
Aya loads governed context from prior work—not from a blank chat.
OpenAya / Product chapter 00
The AI operating system where your work does not disappear into the chat.
A working browser beta of an AI operating layer built to preserve context, act with permission, keep its work inspectable, and remain available wherever the web runs.
01 / The system
Three claims. Each one has proof.
02 / A complete workflow
Recall the work. Use the tool. Keep the state.
The value is not a larger prompt box. It is the continuity between intent, action, and an inspectable durable record.
02
Act
The request moves through the right application, model route, or MCP tool.
03
Write
The result is validated and written as canonical machine-readable state.
04
Rebuild
Indexes and embeddings can be recreated from the files you keep.
03 / Your durable asset
Aya is the OS. Your storage is the disk.
Supabase Storage is the current private reference disk. The contract is provider-neutral so user-owned external disks can follow without changing the file format.
Aya operating layer
Memory governance, model routing, tools, workflows, and user interface.
Mounted disk
Canonical files live under a versioned /aya directory on the selected provider.
Open contract
JSON Schema validation makes the same state legible to Aya and external applications.
04 / Receipts, not claims
If the evidence is absent, Aya should say so.
Public evaluations, capability status, audit aggregates, and machine-readable specifications make the operating layer inspectable.
05 / Plain-language terms
Know what moves, what costs, and what remains yours.
Ownership
Your canonical files, workflow definitions, exports, and generated artifacts remain yours.
Processing
Configured model providers process only the requests needed for the workflow, under their terms.
Costs
Third-party model and storage usage is disclosed; no hidden outcome fee is implied.
Exit
Export or delete your state. Portable files are designed to outlive an Aya subscription.
DESIGN PARTNER MODEL
One named workflow, one shared baseline, one auditable definition of success.
Fixed pilot with an optional outcome credit after mutually agreed success criteria are met. You retain canonical files, exports, workflow definitions, and generated artifacts.
Read the pilot terms06 / Questions
The serious questions deserve direct answers.
What does Aya store?
Aya stores the context, preferences, memories, conversations, projects, decisions, and skills needed for the workflows you choose. The durable representation is versioned, machine-readable files.
Where do those files live?
Today, the reference disk is a private Supabase Storage bucket. Google Drive, OneDrive, Dropbox, and MCP filesystem adapters are planned, not shipped.
Does Aya train foundation models on my data?
OpenAya does not train foundation-model weights on your files. Requests sent to configured model providers are subject to those providers’ processing terms; pilot scopes document the providers involved.
Can I inspect or move my state?
Yes. The disk format has a public specification and JSON Schemas. Canonical files are designed to be readable by Aya and external applications.
What works offline?
The browser shell and selected local functions can operate locally, but model calls, cloud sync, and remote tools require connectivity. Offline-first external-disk sync remains planned.
Can an organization self-host?
The architecture supports bring-your-own infrastructure and portable state. A fully packaged self-hosted enterprise distribution is still beta and should be scoped in a pilot.
Is Aya a hardware product?
No Aya hardware product has been announced. Aya currently begins as a browser-based software beta. Continuity across computers, phones, wearables, and future devices is a device-agnostic research direction, not a release or availability promise.