Pipeline Workflows

Pipelines chain multiple agents together to complete complex tasks.

Creating a Pipeline

pipeline = client.create_pipeline(
    name="code-review-pipeline",
    description="Reviews code, generates fix suggestions, then summarizes",
    steps=[
        {"agent_id": reviewer_id, "step_order": 1},
        {"agent_id": fixer_id, "step_order": 2},
        {"agent_id": summarizer_id, "step_order": 3},
    ],
)

Publishing

Pipelines must be published before other users can run them:

client.publish_pipeline(pipeline["pipeline_id"])

Running a Pipeline

run = client.run_pipeline(
    pipeline["pipeline_id"],
    input={"code": "def hello(): print('world')"},
)
print(f"Pipeline run started: {run['run_id']}")

Monitoring Progress

import time

while True:
    status = client.get_pipeline_run(run["run_id"])
    print(f"Status: {status['status']}")

    for step in status.get("steps", []):
        print(f"  Step {step['step_order']}: {step['status']}")

    if status["status"] in ("completed", "failed", "cancelled"):
        break
    time.sleep(5)

Pipeline Attestations

Each step in a pipeline produces an attestation with a parent_attestation field linking it to the previous step. This creates a verifiable chain of custody from input to final output.

Step 1 → Attestation A (parent: null)
Step 2 → Attestation B (parent: A)
Step 3 → Attestation C (parent: B)