How to Use AI Agent Studio in Sparkforge

Agent Studio is the no-code/low-code workspace in Sparkforge for designing AI agents. Build conversation flows, wire in APIs, attach knowledge bases, and ship the agent to a widget, workflow, or external caller. Agent Studio is available to Agency Admins and Users, and Sub-account Admins and Users.

What you get

• Drag-and-drop builder. Visual canvas for connecting nodes and flow conditions.
• Tool nodes. Pre-built blocks for LLM calls, APIs, knowledge, web search, and the Model Context Protocol.
• Logic and flow control. Branching, conditions, runtime variables, and global variables.
• Version control. Develop, test, and deploy across structured lifecycle stages.
• Built-in testing. Run an agent end-to-end before promoting to production.
• Export and import. Move agents between accounts or back them up.
• API access. Trigger agents programmatically from external systems.

Tool nodes

LLM Node

Calls a large language model to generate or transform text. Configure the prompt, model parameters, and how the response is mapped to downstream variables.

MCP Tool Node

Connects to external tools through the Model Context Protocol. Useful when the agent needs to read data or trigger actions in systems that expose MCP servers.

API Tool Node

Sends and receives data through any HTTP API. Use this to read CRM records, hit a third-party service, or push data into another platform.

Knowledge Base Tool Node

Attaches a knowledge base of documents or stored content so the agent can answer from your own source material instead of generic LLM knowledge.

Web Search Tool Node

Lets the agent pull current information from the public web for time-sensitive answers.

Logic and flow control

• Branching. Route the conversation based on the user's input or earlier tool output.
• Conditions. Gate steps by checking values, variables, or response status.
• Variables. Hold runtime values inside a single agent run, or global values that persist across runs.

Build, test, and deploy

1. Open AI Employee → Agent Studio and click New Agent.
2. Drag tool nodes onto the canvas and connect them to define the flow.
3. Configure each node — prompt for the LLM node, endpoint and headers for the API node, source documents for the knowledge base node.
4. Use the built-in tester to run the agent against sample inputs.
5. Promote the agent through versions until it's production-ready.
6. Deploy to a chat widget, workflow step, or expose via API for external callers.

Where Agent Studio fits

• Customer support agents that read your help docs and hand off to a human when needed.
• Lead qualification agents that score and route inbound chats.
• Operations agents that pull from internal APIs to answer staff questions.
• Voice agents wired into the Voice AI Chat Widget.