Zephr Docs

Architecture

Zephr is a memory and evidence system for coding agents. Everything in this section explains the circuit those capabilities share: who is allowed in, what runs on your machine versus the cloud plane, and where data lives and how it is destroyed.

These pages are the canonical shared pages for the circuit. Capability guides link here instead of restating how identity, persistence, or the boundary work.

Read in order

  1. The product circuit — the whole system on one page: components, capability layers, and what ships today.
  2. Identity and admission — how a connection proves who it is and what it may touch.
  3. Local and cloud — what runs where, and what crosses the boundary.
  4. Persistence — storage planes, encryption, retention, erasure, and migrations.

How status is labeled

Every page distinguishes three things, and never blurs them: shipped (behavior exists in source you can read, with tests), designed (specified in the baseline capabilities, not yet implemented), and planned (named as future work, with no implementation). The status vocabulary is closed; pages carry their status in the badge under the title.

Where the conceptual model lives

The trust and memory vocabulary — beliefs, provenance, review states, the Trust Firewall — is covered in Concepts. This section covers the machinery: the components, boundaries, and lifecycle those concepts run on.