Swarm Builder Guide: Orchestrate Multi-Agent AI Teams on Vincony
# Swarm Builder Guide: Orchestrate Multi-Agent AI Teams on Vincony
Most complex tasks don't belong to a single AI model — they belong to a team. A product launch needs research, copywriting, technical architecture, and a critical review pass. A content campaign needs a strategist, a writer, an SEO specialist, and a visual director. Running those tasks one at a time through a single prompt is slow, and the outputs rarely build on each other in a meaningful way.
That's the problem Swarm Builder solves. It lets you compose a team of specialized AI agents, define how their outputs flow into one another, and run the whole pipeline in one go — watching each agent's activity live as it happens. The result is a set of structured deliverables produced by agents that actually pass context between themselves, not a collection of disconnected responses.
This guide walks through every step of building and running a swarm on Vincony, based on the real UI and controls available in the feature today.
---
How It Works
A swarm is made up of three building blocks:
- Roles — individual AI agents, each with a name, a domain specialty (code, creative, reasoning, search, data analysis, and more), a system prompt, and an assigned model.
- Edges — the workflow connections that determine which agent's output feeds into which agent's input. The Data Flow visualization on the Team Builder tab shows these connections as labeled arrows.
- A project brief — the shared context that all agents receive, describing what the team is collectively trying to accomplish.
When you run a swarm, Vincony saves the current configuration, sends the brief and any additional run-time input to the first agents in the chain, and passes their outputs downstream according to the edges you've defined. The Agent Activity feed on the Run & Monitor tab shows messages from each agent in real time. When the run completes, any structured outputs appear in the Deliverables panel.
Credit cost scales with team size: each role costs 1 credit per run. A five-agent swarm costs 5 credits. The credit badge in the header and on the Run Swarm button always shows the total before you commit.
---
Step-by-Step Walkthrough
Step 1 — Create a New Swarm
Open Swarm Builder. At the top of the list view you'll see the New Swarm card with two fields:
- Swarm name — a short label for your own reference, e.g. "SaaS Launch Team" or "Q3 Content Pipeline."
- Project brief — a plain-language description of what this AI team should accomplish. This is the most important input you'll write: the more specific and outcome-focused it is, the better the auto-generated team and the run results will be.
Fill in both fields and click Create Swarm. Vincony saves the swarm and immediately opens the editor.
Step 2 — Choose a Template or Auto-Build Your Team
Once inside the editor, you're on the Project Brief tab. You have two fast paths to a working team:
Quick Start Templates appear below the brief field. The current built-in templates are:
| Template | Category | Agents |
|---|---|---|
| SaaS Launch Kit | Product | Market Researcher, Copywriter, Design Director, Tech Architect, Fact Checker |
| Content Campaign | Marketing | Content Researcher, Content Writer, SEO Specialist, Visual Director |
| Code Review Pipeline | Engineering | Systems Architect, Senior Developer, Security Reviewer, QA Engineer |
Click any template card to load its roles, workflow edges, and a default brief into your swarm instantly.
If your use case doesn't match a template, click Auto-Build Team. Vincony reads your project brief and uses AI to generate a recommended set of roles with appropriate domains, system prompts, and a sequential workflow connecting them. It's the fastest way to get from a blank page to a working team — and you can always edit the roles afterward.
Step 3 — Review and Refine Roles
Switch to the Team Builder tab. You'll see all agent roles as cards in a grid. Each card shows the role's name, domain, assigned model, and color.
To edit a role, click its card and the Edit Role dialog opens. The configurable fields are:
- Role Name — what this agent is called, e.g. "Lead Researcher" or "Security Reviewer."
- Domain — the specialty used to route model selection. Options include General, Code, Reasoning, Creative, Search, Multilingual, Legal, Math, Data Analysis, Summarization, RAG / Knowledge, Image, Medical, and Vision.
- Output Type — what this agent produces: text, image, audio, code, slides, or data. Setting this correctly helps downstream agents understand what they're receiving.
- AI Model — pick from the full model roster (GPT-5, GPT-5.2, GPT-5.2 Codex, O3 Pro, Claude Sonnet 4.5, Claude Opus 4.5, Gemini 3 Pro, Gemini 3 Flash, DeepSeek V3.2, Sonar Pro, Mistral Large) or leave it on Auto to let Vincony select the best model for the domain. See Models for a capability comparison.
- System Prompt — the per-agent instruction set. This is where you tell the agent its persona, what inputs to expect, and exactly what to produce. Be explicit.
- Color — a visual identifier used throughout the editor and the live activity feed.
To add a completely new role manually, click Add Role in the top-right of the Team Builder tab.
Below the role grid, the Data Flow panel shows every edge as a labeled arrow: "Market Researcher → Copywriter," for example. If you need to change the flow, edit roles and re-run Auto-Build Team, or remove a role and re-add it in the order you want — the auto-build always generates a sequential chain matching the current role order.
Step 4 — Save Your Configuration
Click Save (top-right of the editor header) before running. The save button persists the name, brief, all roles, and all workflow edges to your account. Vincony also auto-saves before every run, so if you forget, the latest state is always used.
Step 5 — Run the Swarm
Switch to the Run & Monitor tab. There's an optional text area for additional run-time input — use this to provide a specific task, a URL, a document paste, or any context that's specific to this particular execution rather than the swarm's permanent setup. If you leave it blank, the project brief is used as the run input.
Click Run Swarm. Vincony saves the swarm, dispatches the pipeline, and starts streaming agent activity into the Agent Activity feed below. Each message is color-coded by the role that produced it, so you can follow the chain as it executes.
If the run produces structured outputs — code files, articles, specs, audit reports — they appear in the Deliverables section once the run completes.
Step 6 — Review History and Deliverables
The Recent Runs row above the activity feed shows all previous executions of this swarm, labeled by timestamp and status (running, completed, failed). Click any past run to reload its activity feed and deliverables. This makes it straightforward to compare runs or retrieve an earlier output.
---
Tips and Best Practices
Write a specific project brief. Vague briefs like "help with marketing" produce generic teams. Specific briefs like "Plan a go-to-market strategy for a B2B SaaS tool targeting HR teams at 50-500 person companies, including competitor analysis, landing page copy, and a 90-day launch roadmap" produce teams calibrated to your actual deliverables.
Match the model to the role's work. Code-heavy roles benefit from GPT-5.2 Codex. Reasoning-intensive roles (fact-checking, legal analysis, security audits) work well with O3 Pro. Creative and long-form writing roles shine with Claude Opus 4.5 or Claude Sonnet 4.5. Leaving a role on Auto is a sensible default, but explicit selection gives you more control over cost and output style.
Use system prompts to specify output format. If a downstream agent needs structured input (a JSON spec, a bulleted list, a markdown document), tell the upstream agent exactly what format to use in its system prompt. This makes the handoff between agents cleaner.
Keep the team lean for focused tasks. A three-agent swarm (researcher → writer → reviewer) is faster and cheaper than a six-agent swarm for most content tasks. Add roles only when the additional specialization meaningfully changes the output.
Save templates you build. If you assemble a team that works well for a recurring task type, rename it clearly (e.g. "Weekly SEO Article Pipeline") and keep it in your swarms list. You can re-run it with different run-time input each time without touching the role configuration.
Credit planning. Free-tier users get a fixed monthly credit allowance. Starter and Pro plans increase the pool significantly, and Power and Business plans are designed for teams running swarms regularly at scale. A five-agent swarm running once a day is 150 credits per month — well within Starter. Check pricing to find the right plan for your volume.
---
Related Features
Swarm Builder is one part of Vincony's agentic layer. If you want a single AI agent that browses the web and takes actions autonomously rather than a coordinated team, try the Web Agent. For hands-off recurring tasks — scheduled content generation, automated reporting, or timed outreach — Auto-Pilot runs your prompts on a cron schedule without manual intervention.
---
Get Started
Stay ahead in AI
Get our weekly AI insights — tips, model comparisons, and guides delivered to your inbox.
No spam, unsubscribe anytime.
Open Swarm Builder, create your first swarm, and click Auto-Build Team on a project brief you've been putting off. The full pipeline — research, drafting, review, and deliverables — can finish in the time it takes to review a single prompt response.
Questions about how swarms fit into your workflow? The help icon on every page links directly to relevant documentation, and the Vincony team is reachable through the platform.