aizk

aizk turns trusted text into scoped, temporal context that any MCP client can recall. It keeps private notes private, lets teams share deliberately, preserves sources and history, and returns one prompt-ready string. One Postgres holds a bi-temporal knowledge graph under forced row level security, and every capability is an MCP tool an agent calls.

Start here

QuickstartConnect a client to the deployment in about five minutes.

ConceptsA plain-language primer for MCP, row level security, and bi-temporal facts.

OnboardingGet an invited collaborator connected with safe memory habits.

MCP clientsPoint Codex, Claude Code, and OpenCode at one OAuth-protected URL.

How it works

The engineHow the store, write path, and read path fit together.

The storeThe content-and-claim union model behind the knowledge graph.

The write pathFrom authored notes and files to normalized Markdown and graph projections.

The read pathHow recall runs the maximal retrieval plan and packs one budgeted response.

Identity and sharingThe Logto boundary and the multi-organization scope model.

The scope-set latticeThe intersection lattice that governs shared and private access.

Reference and operations

APIThe four public MCP tools and the SSH-only maintenance surface.

Architecture rulesThe rules and import contracts that define how aizk is put together.

BenchmarksThree evaluation levels, from unit tests to external benchmarks.

ComparisonHow aizk compares to vault search and other memory systems.

OperationsRun aizk as durable shared infrastructure with backups and upgrades.

SecurityWhat a production deployment protects, trusts, and gates on.