Introduction

Documentation Index

Fetch the complete documentation index at: https://docs.myme.so/llms.txt

Use this file to discover all available pages before exploring further.

Marfa is a typed data layer for structured personal data. It stores items against a shared type system, exposes them through a REST + SSE API, and lets many apps read and write the same person’s data without speaking custom-to-each-other.

​

The problem

Every app has its own data silo. A notes app stores notes one way. A bookmarks app stores bookmarks another. A photo library does its own thing. If you want to use your data across apps — search your highlights alongside your notes, see bookmarks in a timeline with photos — you’re writing glue between incompatible APIs. Marfa sits underneath. One place where typed data from many sources lives together, stays queryable across apps, and carries its provenance honestly.

​

Three primitives

Marfa is built on three primitives. Alongside those, every item carries a tier (library or feed — curated vs. firehose), a lifecycle state (active / archived / trashed), and a provenance stamp (which app wrote it, what kind of actor produced it).

​

How apps use Marfa

Both patterns use the same SDK, the same types, the same wire shape.

​

Design principles

Four principles shape the product. Every other decision flows from them.

• Types are the primary structure. Everything is a typed item. The core type set is opinionated and narrow; custom types are the main road; inheritance lets custom types extend core ones cleanly. See Types.
• Tier is the primary axis of intent. Personal data divides cleanly into a curated library (deliberate, low-volume) and a high-volume feed (low-intent). Every item carries tier: 'library' | 'feed'; storage tiering and permissions are tier-aware. Tier is a prominence signal, not a retention policy — items persist until the user (or an app acting on their behalf) trashes them. See Tier.
• Relationships are first-class and typed. Items connect through named, typed edges — not embedded ids in properties. Edges are queryable from both ends and carry per-type constraints. See Edges.
• Multi-tenant, multi-app, honest provenance. Every item carries source (which app wrote it). In server-synced mode it’s credential-stamped and non-forgeable. See Provenance.

​

What Marfa is not

Scope discipline. A few things Marfa deliberately does not do:

• Not a real-time collaborative editor. Sync is event-driven at the item level, not operation-level. No CRDT.
• Not a binary-replication transport. Moving bytes between devices is for Syncthing, iCloud, Dropbox, Obsidian Sync. The Sync Agent uses those underneath; Marfa provides the typed structure on top.
• Not a federated identity service. Marfa issues OAuth and OIDC identity tokens to apps that trust Marfa as a tenant boundary. It does not resolve user identity across other companies’ systems the way WorkOS, Auth0, or Okta do.
• Not a workflow engine. No rules, automations, or triggers beyond webhooks. Apps build rule engines on top of Marfa’s events.
• Not a protocol or standard. A product with an SDK and a server. The conformance suite checks contract alignment for any reimplementation — Marfa is not trying to be IETF.
• Not a secrets manager. Credentials, passwords, 2FA secrets belong in a password manager or OS keychain.
• No semantic or vector search. Full-text only.
• Not a hosted multi-tenant product at scale yet. Self-hosted by design today.

​

What’s in these docs