EdgeCoder Docs

Appearance

Architecture Deep Dive ​

This page expands the high-level guide into a system-level architecture view. It is derived from EDGECODER_PLAN.md, README.dev.md, and operations/security docs.

Core Services and Responsibilities ​

Agent Runtime ​

  • Two primary paths:
    • interactive path (developer-facing, local-first)
    • worker path (headless swarm participation)
  • Uses execution + model provider contracts shared across modes.
  • Handles plan/code/test/iterate workflow for coding tasks.

Executor ​

  • Enforces allowed language/runtime subset boundaries.
  • Provides controlled execution for generated code paths.
  • Feeds execution/test results back into the agent loop.

Coordinator ​

  • Orchestration-only role for mesh execution.
  • Assigns tasks by capacity/policy, tracks health, and aggregates results.
  • Keeps model access mediated rather than exposing direct public model APIs.

Inference Service ​

  • Split from coordinator to isolate decomposition/inference scaling.
  • Receives decomposition/model requests from coordinator workflows.
  • Can enforce token-gated access in production setups.

Control Plane ​

  • Exposes operator APIs for:
    • network mode
    • rollout policy
    • approvals
    • security/blacklist workflows
  • Serves as a governance/control boundary rather than execution boundary.

Portal ​

  • User identity and account lifecycle:
    • signup/login
    • OAuth
    • passkey support
  • Node enrollment, wallet onboarding, credits and operations views.

High-Level Topology ​

Control Plane vs Runtime Plane ​

Runtime Modes ​

Local-First Mode ​

  • Work stays local when policy requires private-only execution.
  • Swarm coordination is optional.
  • Best fit for regulated or sensitive code paths.

Public Mesh Mode ​

  • Work can be distributed to approved external nodes.
  • Coordinator tracks capacity, quality, and eligibility.
  • Contribution and demand signals feed issuance/economy layers.

Enterprise Overlay Mode ​

  • Adds stricter controls around identity, trust, and boundary enforcement.
  • Designed for higher-assurance environments and policy-heavy operators.

Design Constraints Carried Through the Stack ​

  • Separation of concerns: coordinator vs inference vs control plane vs portal.
  • Explicit trust boundaries: auth, routing, runtime, and economy controls.
  • Auditable state: ledger verification and tamper-evident records.
  • Deployable independence: services can scale and roll independently.

Source Layout Mapping ​

From README.dev.md:

  • src/agent: interactive + worker loops
  • src/executor: subset validation and execution helpers
  • src/swarm: coordinator + worker runtime
  • src/inference: decomposition/inference service
  • src/control-plane: admin and deployment APIs
  • src/apps/ide: IDE provider endpoint
  • src/mesh: peer registration/gossip/protocol handling
  • src/security: blacklist and abuse-control flows
  • src/credits, src/economy, src/ledger: accounting and verification primitives