Getting Started

Build a people research agent in 10 minutes. It takes a name through a chat interface, searches the web with AI search tools, and returns structured JSON with the person's location, profession, and education.

Tutorial

In the sidebar, click the + next to Workflows and name the new workflow "Getting Started". The Start block is already on the canvas — it's the entry point that receives your chat input.

Drag an Agent block onto the canvas and configure its Messages:

System: "You are a people research agent. When given a person's name, use your search tools to find their location, profession, educational background, and other relevant details."
User: insert <start.input> so the agent reads whatever the chat receives.

Leave the Model on the default (claude-sonnet-4-6), or pick any other.

In the Agent block's Tools section, add Exa and Linkup. The agent decides when to call them.

Open the Chat panel, select agent1.content as the output to display, and send: "John is a software engineer from San Francisco who studied Computer Science at Stanford University."

The agent searches, then returns what it found.

If the run doesn't go green, open the output dropdown and read what each block returned — the failure is almost always visible there. The two most common causes are a missing API key on a tool and a reference pointing at a value that isn't there. For the full method, see tracing a failure backward.

In the Agent block's Response Format, click the magic wand (✨) and prompt: "create a schema named person, that contains location, profession, and education". The schema is generated for you.

Back in the Chat panel, select the new structured output option and send: "Sarah is a marketing manager from New York who has an MBA from Harvard Business School."

The agent now returns JSON with location, profession, and education fields — ready for any downstream block to read by name.

Workflows

The full model: blocks, connections, triggers, and how runs execute.

Agent block

Everything the block you just used can do — tools, memory, skills, structured output.

Integrations

1,000+ services your agents can act on, from Gmail to HubSpot.

Deploy it

Ship the workflow as an API, a chat, or an MCP server.

Common Questions

About 10 minutes. You will have a fully working workflow by the end.

No. On hosted Sim, the model and the search tools run on Sim's hosted keys, included with your plan credits. You can bring your own provider keys if you prefer.

No. The whole tutorial uses the visual builder and the chat panel — no code.

Yes. The Agent block supports models from OpenAI, Anthropic, Google, Groq, Cerebras, DeepSeek, xAI, and more, plus local models through Ollama if you self-host.

Not directly. But you can describe what you want to Mothership in natural language and have it build the workflow for you, which is usually faster than manual recreation.

Test iteratively in the Chat panel and inspect each block's output from the dropdown to pinpoint where things go wrong. The Logs page records every run, block by block, with inputs, outputs, timing, and cost.