Skip to content
Control Center

Build your first pipeline

What you will build

Section titled “What you will build”

By the end of this tutorial, you will have:

  • A pipeline template with three steps: trigger → agent task → bash verification
  • A manual run of that pipeline
  • An understanding of how pipelines connect agents, scripts, and conditional logic

Prerequisites: Dispatch your first agent completed.

What are pipelines?

Section titled “What are pipelines?”

Pipelines are DAG-based workflows that chain steps together. Each step has a kind (prompt an agent, run a script, dispatch reviewers, route conditionally, join parallel branches) and can pass data to downstream steps via state.

Templates define the graph; runs are executions. Triggers can auto-start pipelines on domain events or schedules.

For background, see Pipelines and automation.

Step 1: Create a pipeline template

Section titled “Step 1: Create a pipeline template”
  1. Navigate to Pipelines in the sidebar
  2. Click Create template
  3. Name it “code-quality-check”
  4. Add a description: “Run code quality checks on a branch”

The template editor opens with a blank canvas.

Step 2: Add a trigger step

Section titled “Step 2: Add a trigger step”

Every pipeline starts with a Trigger step. This is the entry point.

  1. The canvas already has a trigger node. Click it.
  2. Leave the default configuration — it accepts manual runs.

Step 3: Add an agent step

Section titled “Step 3: Add an agent step”
  1. Click Add step on the canvas
  2. Select Prompt agent as the step kind
  3. Configure it:
    • Agent: pick your coder agent
    • Prompt: “Review the code in the current branch for common quality issues. List any problems you find.”
    • Mode: chat
  4. Connect the trigger step to this agent step (drag from trigger’s output to agent’s input)

Step 4: Add a bash script step

Section titled “Step 4: Add a bash script step”
  1. Add another step, kind Bash script
  2. Configure it:
    • Script: echo "Quality check complete"
    • This is a simple verification step — in practice you’d run linting or test commands
  3. Connect the agent step to this script step

Your pipeline now has three connected steps: trigger → agent → script.

Step 5: Run the pipeline

Section titled “Step 5: Run the pipeline”
  1. Click Run pipeline in the header
  2. If the template declares inputs, fill them in (our template has none)
  3. Click Start run

The pipeline engine:

  1. Creates a PipelineRun record
  2. Starts the trigger step
  3. When the trigger completes, schedules the agent step
  4. The agent dispatches, works, and completes
  5. When the agent step completes, schedules the script step
  6. The script runs and completes
  7. The run status moves to Completed

Watch the run progress in the pipeline run detail view. Each step shows its status, input, and output.

Step 6: Inspect the results

Section titled “Step 6: Inspect the results”

Click any step in the run to see:

  • Input: the data passed into the step
  • Output: what the step produced
  • Status: completed, failed, or skipped
  • Duration: how long it took

The agent step’s output contains the agent’s response. The script step’s output contains the stdout.

Recap

Section titled “Recap”

You created a three-step pipeline, ran it manually, and inspected the results. Pipelines can be much more complex — conditional routing, parallel branches, join nodes, error handling — but the basic pattern is the same: steps connected in a DAG, passing data through shared state.

Next steps: