Docs
Start building
Hiring Your Crew

Configuring nodes

The inspector: giving each node its instructions, tools, and settings.

2 min read · 3 sections
Part of the crew rollout: Agents shows as Soon in the app until it reaches your account. See Hiring your crew.

A node on the canvas is a step; the inspector is where you say what that step actually does. Select any node and a panel opens on the right with the settings for that node's type.

The agent node

The thinking core of most employees. Its config is where the real behavior lives:

  • Instructions: the system prompt (the employee's job, voice, and rules).
  • Model: which LLM runs the node.
  • Tools: the built-in capabilities it can call (web search, HTTP, memory, extended thinking, on-chain actions, and your connected integrations).
  • Memory: whether it remembers the conversation, and how far back.
  • Temperature & max steps: how creative it is, and how many tool-loops it may run before it must answer.

Tools an agent can use

Tools are how an employee does things rather than just talk. They resolve by name:

ToolWhat it does
web_searchSearch the web.
http_requestCall any HTTP endpoint.
IntegrationsActions on connected apps: Slack, GitHub, and more, via OAuth.
GatewayThe web3 providers behind Modus Gateway: chain reads, prices, routing.
On-chain actionsWallet and contract operations.
Memory filesBuilt-in filesystem tools read and write the agent's long-term memory files.
thinkExtended reasoning before answering.

Other node configs

  • App: pick the integration and the specific action, then map your data into its fields. Connecting an app runs an OAuth flow.
  • HTTP: set the URL, method, headers, body, and auth.
  • Branch: name each case and give it a condition; each case becomes its own output path.
  • Memory: choose the history window.
  • Prompt template: write a template with {{variables}} that pull from earlier nodes.