Why toda

The context problem

Every conversation with a model produces context worth keeping. Preferences, decisions, the state of the thing you are working on. That context is what makes an assistant feel like it knows you.

Right now it is stranded. Each app keeps its own memory, in its own format, behind its own login. The moment you leave, it is gone. A company running five different models maintains five disconnected copies of the same knowledge. A developer building on top of any of them rebuilds storage and retrieval from scratch. Agents that could share what they learn instead each start blind.

The data is being created. Nobody owns it, and it does not move.

Memory that belongs to you

toda separates memory from the app that captured it.

A memory is written once and read back anywhere: across products, across models, across autonomous systems. The same layer holds personal context, company knowledge, shared datasets, and agent state. Ownership sits with a wallet instead of an app account, so the memory follows the person, not the product.

Keeping it honest

An open memory layer invites junk. If anything can write, most of what gets written is duplicate, low value, or deliberately poisoned.

toda keeps the network honest through validation. An AI layer reads each incoming sequence and keeps only what carries durable signal, a real decision, a stable preference, a fact worth holding, while dropping noise, spam, and content shaped to poison a model. Stakers drive that process. They act as the network's validators, and their stake is what puts the AI to work over a given sequence, so keeping memory has a cost and flooding the network with junk has a bigger one. Stake reflects conviction that something is worth keeping, not proof that a fact is true.

Private by construction

Raw content never touches the chain. It is stored offchain and encrypted under keys bound to the owner's wallet. What goes onchain is a cryptographic commitment, enough to prove a memory is yours and has not been altered, and nothing of the content itself.

That split is the point. Private data gets to participate in a public, verifiable network without being exposed by it.

Next: How it works.