Skip to content

AIgency architecture

AIgency is a scheduling layer over ordinary sessions — not a second execution engine or a parallel set of provider transports.

AgentDock Pro

AIgency is available only in AgentDock Pro.

The single most important architectural fact about AIgency: it does not introduce a new way of running an agent. Every task it executes is a real AgentDock session, created through the exact same session-service a normal hand-typed conversation uses — the same adapters, the same permission modes, the same conversation view.

The pieces

  • Planner (aigency-planner-service.ts) — turns an objective into tasks, dependencies, and draft acceptance criteria. See Planning.
  • Scheduler (aigency-service.ts) — the dependency-aware concurrent scheduler that decides what runs next. See AIgency scheduler.
  • Routing (aigency-agent-routing.ts, aigency-agent-profiles.ts) — capability-matrix scoring for task assignment. See Task routing.
  • Worktree service (aigency-worktree-service.ts) — isolates concurrent writable tasks into git worktrees. See AIgency worktree service.
  • Verification service (aigency-verification-service.ts) — evaluates acceptance criteria against captured evidence. See AIgency verification pipeline.
  • Repositories — durable persistence for runs, tasks, interactions, usage, timeline, criteria, defects, and completion reports. See AIgency persistence.

A ready task is just a normal launch

"Ready" means a task's dependencies are satisfied and concurrency/capacity allow it to start — at that point AIgency launches it exactly the way you'd start any other session, just without a human typing the first message.