DAG Pipeline

The Autonomous Sub-Agent Pipeline runs a goal through a directed acyclic graph of specialized sub-agents — Architect → Planner → Coder → Analyzer → Tester → Security → Docs → Benchmark → Debugger → Stager.

Running a pipeline

1. Open the DAG view (icon rail) or keep it inside the Assistant.
2. Optionally type a goal in the composer first — otherwise a default goal is used.
3. Click Run Sub-Agent DAG.

Each task streams a real AI response from the model assigned to its role (see below), the repository context is compacted once per run and shared by all tasks, and parsed diffs land on the task.

Assigning models per execution mode

The Per-Execution-Mode Model Assignment card in the DAG view lists all 10 roles. Each row is a dropdown of your enabled fallback tiers plus Inherit from Auto Router:

• Pick a tier per role — e.g. Architect on Claude, Tester on the cheapest model.
• Consolidate to one model applies the first row to everything.
• Apply to all on a row spreads that row's model to every other role.

The mapping is logged when the run starts, and each task card shows its model as a green badge.

Watching the run

• The Plan & Tasks tab shows the live plan with LEDs per step.
• The DAG cards update: Pending → Running → Completed / Failed.
• Logs capture every transition, including the model each task ran on.
• A task whose upstream dependency failed is skipped with a clear reason.

Pipeline modes

The Mode button cycles presets: FullAutonomous, StandardEngineering, FastPatch, SecurityAndCompliance, DocumentationAndTests, Debugger, Custom. You can also build a Custom DAG in the editor card and import/export pipelines as JSON.