Documentation

These docs come in two halves. The user half is for anyone whose agent talks to aizk. It explains what memory is, how sharing works, and how to write and ask so the answers stay useful. You need no database knowledge to read it. The developer half is for anyone changing aizk or running a deployment, and it goes all the way down to table columns, job schedules and GPU budgets.

Read any page on its own in ten minutes or less. When a page needs an idea it does not own, it links the one page that does instead of explaining it twice.

Start here

What aizk isThe shortest complete explanation, in plain language.

QuickstartConnect a client and store your first memory in five minutes.

Your first hourA guided walk from an empty memory to a shared, useful one.

GlossaryEvery term these docs use, defined once.

Understand what it does

Sources and derived knowledgeWhy only one of the two kinds of stored thing is authoritative.

ScopesWho can read a memory, expressed as a set rather than a permission list.

Time and historyWhen something was true, when aizk learned it, and how corrections keep the past.

Evidence and provenanceWhat recall gives back, and why it is never an answer.

Build on it

System mapEvery moving part on one interactive diagram.

The data modelThe content and claim union behind the whole store.

How recall runsThe seven steps between a question and its evidence.

Development setupA working checkout, a database and the model lanes.

Run it

Hardware and costWhat it takes to run, and how model choices move that number.

First startFrom an empty host to a working deployment.

The security modelWhat a deployment protects, trusts, and fails closed on.

Backups and recoveryWhat is backed up and what restoring really restores.