DEV Community

Cover image for We built shareable multi-agent workflow templates — here's how they work
Vincent Grobler
Vincent Grobler

Posted on • Originally published at crewform.tech

We built shareable multi-agent workflow templates — here's how they work

Most "AI agent" platforms give you a single agent and a prompt box.

That's fine for simple tasks. But real work — research pipelines, content creation workflows, code review chains — requires multiple agents working together, each with a specific role, wired in sequence.

The problem? Setting that up from scratch every time is tedious.

So we built Workflow Templates.


What are Workflow Templates?

A workflow template is a portable blueprint that packages:

  • Multiple agents (with their roles, prompts, models, and tools)
  • A team configuration (pipeline, orchestrator, or collaboration mode)
  • Variables that get resolved at install time (like {{model}} or {{topic}})

You browse templates in the marketplace, click Install, fill in your variables, and get a fully wired multi-agent team in your workspace — ready to run.

┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  Researcher  │───▶│   Writer    │───▶│   Editor    │
│  (gpt-4o)    │    │  (claude)   │    │  (gpt-4o)   │
└─────────────┘    └─────────────┘    └─────────────┘
        ▲
        │  Install from template
        │  Fill in: topic, model, tone
Enter fullscreen mode Exit fullscreen mode

5 starter templates

We're shipping with five templates to get you started:

Template What it does
Content Research Pipeline Research → Write → Edit. Three agents that turn a topic into a polished article.
Code Review Team Reviewer + Security Auditor. Catches bugs, vulnerabilities, and style issues.
Data Analysis Crew Analyst → Visualizer → Report Writer. Raw data → structured insights.
Customer Support Triage Classifier → Responder → Escalator. Routes and handles support tickets.
SEO Content Factory Keyword Researcher → Writer → SEO Optimizer. End-to-end content production.

Each one is fully customizable after install. Change the model, tweak the prompts, add tools, swap agents — it's your workflow.

Creating your own templates

The Creator UI lets you turn any existing team into a shareable template:

  1. Pick your team — select an existing pipeline/orchestrator/collaboration team
  2. Add variables — mark dynamic parts (model, API key, topic) as {{variables}}
  3. Set metadata — name, description, category, icon
  4. Publish — it appears in the marketplace for your workspace (or the community)

The schema is simple JSON:

{
  "name": "Content Research Pipeline",
  "version": "1.0.0",
  "agents": [
    {
      "name": "Researcher",
      "role": "Research Analyst",
      "model": "{{model}}",
      "system_prompt": "Research {{topic}} thoroughly..."
    }
  ],
  "team": {
    "mode": "pipeline",
    "steps": ["Researcher", "Writer", "Editor"]
  },
  "variables": {
    "model": { "type": "model_select", "default": "gpt-4o" },
    "topic": { "type": "text", "label": "Research Topic" }
  }
}
Enter fullscreen mode Exit fullscreen mode

7-day Team trial for new signups

Starting today, every new CrewForm signup gets 7 days of full Team-tier access — no credit card required:

  • ✅ Orchestrator Mode (brain + workers)
  • ✅ Collaboration Mode (multi-agent discussion)
  • ✅ A2A Protocol (agent-to-agent interoperability)
  • ✅ Team Memory (pgvector semantic search)
  • ✅ RBAC + workspace invitations
  • ✅ Unlimited agents and tasks

After 7 days it reverts to the free plan. No gotchas.

Why we built this

We kept seeing the same pattern: someone would sign up, create one agent, run one task, and leave. Not because CrewForm didn't work — but because single-agent tasks don't show the magic.

The magic is in multi-agent coordination. A researcher that feeds a writer that feeds an editor. A code reviewer that escalates to a security auditor. An orchestrator that delegates to specialists.

Templates + the trial let new users experience that within the first 5 minutes.

Try it


CrewForm is open source (AGPL-3.0), self-hostable, and supports 16+ LLM providers including local models via Ollama. BYOK — we never touch your API spend.

Top comments (0)