Mastra Editorial Pipeline

A distributed editorial pipeline combining Render Workflows for orchestration and Mastra agents for review and revision. The UI starts editorial_pipeline. That parent fans out three review_draft runs, then revise_draft.

What This Demo Shows

This repo demonstrates how to split a Mastra agent pipeline across Render Workflows tasks:

Platform	Role
Render Workflows	Queues each task on its own instance, with retries, timeouts, and parallel fan-out
Mastra	Reviewer and editor agents (openai/gpt-5.6-sol in the Mastra Render guide)
Render Web Services	Express API plus a live UI that streams progress over SSE

How It Works

1. Browser posts a draft to the Express API on Render
2. Express starts editorial_pipeline and streams each chained child run over SSE
3. The parent runs three review_draft tasks in parallel, then revise_draft

Render Workflow Task	Mastra agent	What it does
editorial_pipeline	—	Parent: Promise.all on three reviews, then revise
review_draft	Reviewer	One task run per focus: technical clarity, structure and flow, reader usefulness
revise_draft	Editor	Rewrites the draft from the combined reviews

The live UI starts the same parent the Mastra Render guide starts. Bars and cards are those child task runs.

4. Open the live page, click Run pipeline, and watch the timeline before the rewrite appears.

Quick Start

Prerequisites

• Render account
• An API key from a Mastra model provider (OpenAI, Anthropic, Google, and others)

Deploy

Blueprints cannot create Workflow services yet, so this demo is set up in the dashboard.

1. Create the Workflow service:

   • Go to Render Dashboard → New → Workflow
   • Connect this repository
   • Build command: npm install && npm run build
   • Start command: npm run start:workflows
   • Name: render-workflows-mastra-workflow
   • Add env var: OPENAI_API_KEY

2. Create the web service from render.yaml, or New → Web Service on the same repo:

   • Build command: npm install && npm run build
   • Start command: npm start
   • Add env vars: RENDER_API_KEY (create one) and WORKFLOW_SLUG=render-workflows-mastra-workflow

3. Open the web service URL, load the sample draft, and click Run pipeline

The Render CLI can create the same Workflow service:

./scripts/create-workflow-on-render.sh

Features

Feature	Description
Parallel reviews	Three review_draft runs start together via Promise.all inside editorial_pipeline
Independent retries	A flaky model call retries that review without restarting the others
Per-task compute	Reviewers use starter; the editor uses standard with a longer timeout
Live progress	The UI streams SSE events while the parent task runs

Configuration

Variable	Where	Description
RENDER_API_KEY	Web service	Render API key for dispatching tasks
WORKFLOW_SLUG	Web service	Must match the Workflow slug (render-workflows-mastra-workflow by default)
OPENAI_API_KEY	Workflow service	Required by the agents on the Mastra guide (openai/gpt-5.6-sol)
POLL_INTERVAL_MS	Web service	How often Express polls the parent task (default 800)

Project Structure

src/                         Same files as the Mastra Render guide
  index.ts
  mastra/
  tasks/
web/                         Live UI around those tasks
  server.ts
  pipeline/
  static/index.html
render.yaml                  Web service Blueprint

API Routes

Method	Path	Description
GET	/	Editorial UI
GET	/health	Health check
POST	/review	{ "draft": "..." } → SSE stream (status, done, error)

Troubleshooting

Problem	Solution
Tasks fail immediately	Set OPENAI_API_KEY on the workflow service, not only the web service
startTask returns but nothing runs	WORKFLOW_SLUG must match the Workflow slug in Dashboard → Workflow → General
Empty model output	The task throws and retries; check the model name against Mastra's router
Web service cannot dispatch	Set RENDER_API_KEY on the web service

Learn More

Render:

• Render Workflows Documentation
• Defining tasks
• Render Developers Discord

Mastra:

• Mastra on Render
• Mastra agents
• Model router

License

MIT