Skip to main content
One agent should have one responsibility. When a process involves several intentions, such as qualifying a lead and then collecting a payment, split it into agents and connect them with transfers. Each agent stays short, testable, and easy to fix. This is the multi-agent setup the Nexor team shows at the end of every onboarding.

​
Why split

A single agent with one very long prompt has to juggle every situation at once. Symptoms appear quickly: it skips steps, mixes rules from different flows, and every edit risks breaking something else. Splitting by intention keeps each prompt focused on one outcome, gives each agent its own pipeline, variables, tools, and follow-up rhythm, and lets you test each piece alone. Split when goals, rules, tools, cadence, risk, ownership, or completion criteria differ materially. Keep one agent when they do not.

​
How a transfer works

A transfer is configured on a pipeline stage. When a lead enters that stage, Nexor moves it to the target agent.
Pipeline stage with transfer configuration

A stage configured to transfer the lead to another agent.

  • The conversation history always carries over. The next agent continues the same thread; the lead does not notice a restart.
  • You choose which fields to copy. Variables collected by the first agent can prefill the next one, so nothing is asked twice.
  • The receiving agent starts in its own pipeline with its own rules, tools, and follow-up rhythm.
Other common pairs: a support agent that transfers sales questions to a sales agent, an information agent that transfers ready-to-book leads to a booking agent, and a collections agent that transfers disputes to a human-review stage.

​
Design the handoffs before building

Fill in one row per agent before you create anything. It prevents loops and makes ownership clear.
Rules of thumb:
  • Every lead must end with one agent that owns the final outcome.
  • Transfers go forward. Avoid A → B → A unless there is a clear reason and a stop condition.
  • Name agents by intention: “Qualifier”, “Payment”, “Booking”, not “Agent 2”.

​
Build the system

  • With NexorGPT: describe the flow and the agents you want. NexorGPT proposes the system, creates the agents, and configures the transfer stages. See Build with NexorGPT.
  • In settings: create each agent, then open the stage that should hand off and configure the transfer target and the fields to copy.
  • From the API or MCP: developers can set transfers on a stage programmatically. See Update pipeline structure and the MCP server.

​
Test the whole flow

Open the playground on the first agent and drive the conversation to the transfer condition. Then confirm that the next agent picked up with the history and the copied fields, and that it follows its own rules from there. See Test in the playground.

​
Group agents for reporting

Agents that belong to the same process can be rolled up under a Group, so the dashboard shows one combined figure for the whole flow rather than one per agent. Set it under Roll up reporting under in each agent’s advanced settings. See Agent settings.

​
Next

Hand the technical pieces to your team with For developers, or go to the Launch checklist.
Last modified on September 7, 2026