Tutorial: Build a support assistant

What you'll build

A two-agent pipeline that triages support tickets, queries knowledge bases, and drafts helpful responses in under 30 minutes.

Prerequisites

• Completed the Quickstart
• Knowledge base stored in data/support_articles/
• Provider keys configured in .env

Step 1 — Create the pipeline config

pipelines/configs/support.yaml

agents:
  triage:
    profile: support/triage@v1
  responder:
    profile: support/responder@v1
tools:
  knowledge_base:
    path: contextagent.tools.notebook.VectorSearchTool
    scopes: [support]
  escalation:
    path: contextagent.tools.support.EscalationTool
pipelines:
  support_assistant:
    entrypoint: pipelines.support.SupportPipeline

Step 2 — Author agent profiles

contextagent/profiles/support/triage.yaml

prompt: |
  You are an experienced support triage specialist...
state:
  severity_threshold: 0.8
tools:
  allow: [knowledge_base]

contextagent/profiles/support/responder.yaml

prompt: |
  You craft empathetic, helpful, and concise replies...
tools:
  allow: [knowledge_base, escalation]

Step 3 — Implement the pipeline

from pipelines.base import BasePipeline
from pydantic import BaseModel

class SupportQuery(BaseModel):
    ticket_id: str
    body: str

class SupportPipeline(BasePipeline[SupportQuery]):
    async def run(self, query: SupportQuery):
        triage = await self.context.agents["triage"](body=query.body)
        if triage.escalate:
            return await self.context.tools.escalation(ticket=query.ticket_id, reason=triage.reason)

        docs = await self.context.tools.knowledge_base(
            question=query.body,
            limit=5,
        )
        reply = await self.context.agents["responder"](
            ticket=query.body,
            docs=docs,
        )
        return reply

Step 4 — Add observability

        with self.tracer.span("support.reply", attributes={"ticket": query.ticket_id}):
            ...

Tip

Capture every generated reply as an artifact: self.artifacts.save_text("replies/{query.ticket_id}.md", reply)

Step 5 — Run the tutorial

uv run python -m pipelines.support --ticket-id 1234 --path data/tickets/1234.json

Checkpoint

• ✅ Triage agent labels priority and next action.
• ✅ Responder agent drafts a reply referencing the knowledge base.
• ✅ Escalations route to the escalation tool with metadata.

Next up: polish the experience with Pipeline Orchestration.