On this page

System Context๏ƒ

Architecture and design home

Reader Question๏ƒ

Where does Melder sit relative to application code, operators, agents, and persistence?

Short Answer๏ƒ

Melder runs inside the application's Python process. Application code defines and enters the object world through bind, conjure, and meld. Optional external systems do not own that world: orchestrators or agents enter through bounded Rift surfaces, while external storage carries structural persistence records through user-provided adapters.

Melder system context

Editable diagram source

Boundary Responsibilities๏ƒ

  • Application code owns domain behavior and decides what to register.

  • Melder owns graph compilation, scope/lifetime enforcement, and runtime structure.

  • Developers/operators define configuration and lifecycle policy.

  • Orchestrators/agents remain optional consumers of controlled live-system surfaces.

  • External persistence remains caller-selected infrastructure; Melder records structure without requiring a database dependency.

Why This Design Is Strong๏ƒ

Melder does not replace the application's entry point or domain model. It supplies the systems layer beneath them, keeping domain objects ordinary while making their runtime relationships explicit and inspectable.

Tradeoffs๏ƒ

Embedding the runtime in-process gives direct object access and shared-memory performance. It also means application owners must define process, frame, and cleanup boundaries deliberately instead of delegating them to an external container service.

Where to Go Next๏ƒ

Source entry points:


Canonical source ยท Full contents