Skip to content

Architecture

agentic-proteins architecture is the map of a shrinking compatibility bridge. The point of this section is not to admire the old structure. The point is to show exactly how legacy entrypoints are kept alive while canonical ownership already lives elsewhere.

flowchart LR
    legacy["legacy callers<br/>imports, CLI, API"]
    interfaces["compatibility entrypoints<br/>interfaces"]
    bridge["bridge families<br/>agents, execution, providers, tools"]
    state["bridge state seams<br/>state, artifacts, outputs"]
    canonical["canonical packages<br/>runtime and lower layers"]

    legacy --> interfaces
    interfaces --> bridge
    bridge --> state
    bridge --> canonical
    state --> canonical

What This Diagram Is Saying

  • readers should be able to separate preserved entrypoints from canonical ownership in a few seconds
  • the bridge still has real structure, but that structure exists to forward, constrain, and expose migration proof rather than reclaim authority
  • if a module starts to look like the only place behavior can live, the package is drifting back into permanence

Start With

  • open Execution Model when you need to follow a legacy import, CLI call, or API request until it reaches canonical runtime ownership
  • open Integration Seams when a change risks moving real authority back into the compatibility layer
  • open Module Map when you already know the question and need the owning files quickly

Reading Lenses

Fast Proof Points

  • src/agentic_proteins/interfaces/ for preserved CLI, HTTP, and structure-report entrypoints
  • src/agentic_proteins/agents/, execution/, and tools/ for the bridge families most exposed to accidental shadow-ownership drift
  • src/agentic_proteins/providers/ and state/ for adapter selection and compatibility state that still survive for migration review

Boundary Test

If a reviewer cannot point to the forwarding seam and the canonical owner in the same explanation, the architecture is still hiding the real system.