Pipelines

Automated Pipelines for AI Agent Workflows

Define DAGs of tasks. Agents execute in parallel where possible, in sequence where required. Dependencies auto-resolve. Contracts auto-exchange.

From Tasks to Pipelines

Individual tasks are useful. But the real power comes from chaining tasks into automated pipelines:

Agent A analyzes the codebase and produces a migration plan
Agent B implements database changes based on the plan
Agents C and D update the backend and frontend in parallel
Agent E runs integration tests after both are done
Agent F deploys if all tests pass

Without orchestration, you manually trigger each step. With Orchestratia, you define the pipeline once and let it run.

Pipeline Orchestration with DAGs

Orchestratia's dependency system supports directed acyclic graphs (DAGs) of tasks. Combined with auto-assignment and contract exchange, this creates fully automated pipelines.

graph TB
    A["Analyze Codebase<br/>📋 Migration Plan"] --> B["DB Migration<br/>📋 Schema Changes"]
    A --> C["Backend Update<br/>📋 New API Endpoints"]
    A --> D["Frontend Update<br/>📋 UI Components"]
    B --> C
    C --> E["Integration Tests<br/>📋 Test Report"]
    D --> E
    E --> F["Deploy<br/>📋 Deploy Status"]

    style A fill:#F0F9F7,stroke:#2A9D88,stroke-width:2px
    style B fill:#FAF8F5,stroke:#E8E3DA
    style C fill:#FAF8F5,stroke:#E8E3DA
    style D fill:#FAF8F5,stroke:#E8E3DA
    style E fill:#FAF8F5,stroke:#E8E3DA
    style F fill:#F0F9F7,stroke:#2A9D88,stroke-width:2px

How Pipelines Work

Define tasks with dependencies and structured specs
Auto-assignment places ready tasks on capable servers
Agents execute — some in parallel (C and D), some sequentially (B before C)
Contract exchange — each task's output becomes the next task's input
Cascading resolution — when a task completes, the hub checks what's unblocked and assigns next tasks
Human checkpoints — insert plan mode at critical points for approval before proceeding

Fan-Out / Fan-In Patterns

The DAG supports common orchestration patterns:

Fan-out: One task unblocks multiple parallel tasks (A → B, C, D)
Fan-in: Multiple tasks must complete before one can start (C, D → E)
Linear chain: Sequential execution with data passing (A → B → C)
Diamond: Fan-out then fan-in (A → B, C → D)

DAG Pipelines

Define directed acyclic graphs of tasks for complex multi-step workflows.

Auto-Resolution

Dependencies resolve automatically. Ready tasks assign to capable servers instantly.

Contract Passing

Each task's structured output feeds into downstream tasks as resolved inputs.

Human Checkpoints

Insert plan mode at critical pipeline stages for human approval before proceeding.

Build your first agent pipeline

Define tasks with dependencies and let Orchestratia automate the execution flow.

Get Started View Documentation