What is AIgency?

AIgency lets you describe an objective once and have AgentDock break it into tasks, assign each task to a coding agent, and run them — potentially several at a time — until the objective is met.

It orchestrates ordinary sessions

AIgency is not a separate execution engine. Every task it runs spawns a real AgentDock session through the same session service used by a normal, hand-typed conversation — the same Claude Code, Codex, and Antigravity adapters, the same permission modes, the same conversation view. AIgency's job is purely the scheduling layer on top: deciding what runs, in what order, on which agent, and with how much concurrency.

Runs and tasks

A run holds the objective, constraints, and completion criteria for a piece of work, plus the list of tasks generated to satisfy it. See Creating a run.

A task is one unit of work assigned to one agent. Tasks can depend on each other, forming a dependency graph, and each has a kind — standard, integration, or verification — that affects how and where it executes. See Execution.

What AIgency actually does

• Creating a run — define an objective and let AgentDock plan the work.
• Multi-agent orchestration — the concurrent scheduler, concurrency limits, and failure handling.
• Agent assignment and task routing — which agent runs which task, and why.
• Execution — task statuses, isolated git worktrees, and shared-workspace fallback.
• Approvals — permission modes at the run and per-agent level, plus a final human sign-off.
• Retries — automatic retry and capacity-aware backoff.
• Run history — the activity timeline, defects, reports, and reusable templates.