Building Multi-Agent Systems in Captivation Hub Agent Studio

Agent Studio is the visual, no-code/low-code builder inside Captivation Hub for designing AI-driven multi-agent systems. With drag-and-drop nodes, conditional routing, and built-in testing, you can launch sophisticated workflows that blend autonomous AI reasoning with deterministic logic — without writing a single line of code.

What Is Agent Studio?

Agent Studio is a canvas-style builder that lets you orchestrate multiple AI agents and rule-based steps inside one workflow. Each agent can tap into LLMs, knowledge bases, web search, third-party APIs, and more, while sequential nodes handle predictable tasks like validation or form capture. The result is a single, unified automation that can think, decide, and act on your behalf.

Visual Agent Builder

The canvas gives you an intuitive way to map out entire workflows: drag nodes onto the grid, connect them with edges, and rename or reorder anything via right-click context menus. Cleaner node layouts and updated connection lines make even large systems easy to follow at a glance.

Node Types

Agent Studio ships with two core nodes you can combine freely:

• AI Agent Node — connect LLM prompts to tools like Web Search, Knowledge Base Search, GenAI, or external APIs for dynamic reasoning, content generation, and data look-ups.
• Sequential Node — perform rule-based actions such as API calls, form validation, or webhook triggers when you need absolute control and repeatability.

Conditional Routing

Instead of separate "router" blocks, the edges themselves hold the logic. Write simple expressions (e.g., orderTotal > 500) or use AI intent matching to decide which downstream node should fire. This keeps your flows compact while still allowing highly granular branching.

Knowledge Base for RAG

Attach multiple knowledge sources — PDFs, CSVs, FAQs, or full websites — to any AI Agent Node. Retrieval-augmented generation (RAG) keeps answers grounded in your own content while still benefiting from the LLM’s creativity. You can even set source-level filters for surgical precision.

Global Variables and Runtime Inputs

Store API keys, environment constants, or branding guidelines once, then reference them anywhere in the workflow. Pass runtime variables (e.g., locationId, userId) from the calling app or chat session so every execution feels personal.

Version Control

Every save creates a draft. When you’re ready, publish to Production with one click — or instantly roll back to any prior version if something breaks. No separate Git flow required.

Testing and Deployment

The Test panel lets you run the entire agent — including LLM calls and tool invocations — before a single end-user ever sees it. Once verified, deploy to Ask AI immediately. Additional channels like web widgets and API endpoints are coming soon.

Step 1 — Set Up Agent Studio

Building your first multi-agent workflow is straightforward. Here’s the full walkthrough.

Navigate to Agent Studio

From your Captivation Hub dashboard, click AI Agents in the left sidebar, then select the Agent Studio tab. You’ll land on the Agent Studio dashboard, where all of your agents are listed.

Create a New Agent

Click + Create Agent in the top-right corner. This opens the canvas, your workspace for building workflows.

Here’s what you’ll see:

• Left panel — available nodes (your building blocks).
• Center — the canvas where you build the workflow.
• Right panel — settings, where you configure each step.

Click Create Agent to open a blank canvas.

Add an AI Agent Node

AI Agent Nodes are the core of your workflow. They let your agent think, generate responses, and use tools like Knowledge Base or Web Search. From the left panel, drag AI Agent onto the canvas. This node will handle conversations and decision-making.

Configure the AI Agent

Click on the AI Agent node to open its settings panel and fill in:

• Model — select the recommended model (for example, GPT-4.1).
• Prompt — define how the agent should behave.

Example prompt:

You are a helpful assistant for a business. Answer user questions clearly and politely. If you don’t know the answer, ask for more details.

The prompt acts as the "instructions" for your AI agent — it directly shapes how the agent responds.

Connect the Flow

Workflows run from one step to the next using connections. Click and drag from the small connector dot on a node and connect it to the next step (or leave it as a single-node flow for now). If no condition is added, the flow simply continues to the next step automatically.

Add Sequential Nodes (Optional)

Sequential Nodes are used for structured, rule-based actions such as API calls or validations. Drag a Sequential Node from the left panel any time you need that kind of deterministic logic.

Add Variables (Optional but Powerful)

Variables let you reuse values like names, IDs, or API keys across the workflow. Click Variables at the top of the screen and add a new variable under the Global tab.

A few common use cases:

• Storing API keys
• Passing user-specific data
• Personalizing responses

Test Your Agent

Testing makes sure your agent behaves correctly before going live. Click the Test button at the top, enter a sample question (e.g., "What services do you offer?"), and review:

• Response quality
• Tool usage (if added)
• Flow behavior

Publish Your Agent

Once you’re happy with the results, click Publish → Production.

Frequently Asked Questions

Who can access Agent Studio in my account?

Agency Admins/Users and Sub-account Admins/Users with the right role permissions can build and manage agents.

Does every edge need a condition?

No — if you omit a condition, the edge simply behaves like a default "next" step.

Can one agent call another agent?

Yes. Place the second agent inside an AI Agent Node as a sub-workflow, or trigger it via API from a Sequential Node.

How do I track LLM token usage?

Usage shows up under Settings → Billing → AI Usage once the agent runs in Production.

Is data from Knowledge Bases sent to the LLM?

Only the retrieved snippets are passed — not the entire documents — which keeps context tight and costs low.

What happens if an external API fails inside a Sequential Node?

The error is surfaced in the run log, and you can branch to a fallback node using conditional edges (for example, on specific error codes).

Can I export an agent to another account?

Export/import functionality is on the roadmap. For now, rebuild or duplicate the agent manually within the same account.

Are multiple versions of one agent billable separately?

No — only Production runs consume tokens. Drafts and tests use your sandbox quota.