Multiple agents that collaborate
One agent, one trade. When something else is needed, the agent asks a colleague or hands the conversation over.
Specific instructions and a few coherent tools for each one.
A fast model on the front line and a more capable one for hard work.
Delegate or hand off
| Port | Tool | What happens |
|---|---|---|
| Delegate a task | assignTask | The agent gives a task to a colleague in a new conversation and waits for the reply (by default up to 600 seconds), or leaves it running in the background |
| Hand off | handoff | The whole conversation passes to the other agent, which replies from that moment on |
- The description written on each port is what the model reads about the agent at the other end: it decides when to delegate.
- With Inherit User the delegated conversation belongs to the same user, and the colleague knows who it is working for.
- To add a collaborator you add one port per agent, each with its own description.
Subtasks
An agent can hand a piece of work to an empty copy of itself, with the same instructions and tools but without the conversation's context. It is on by default and can be turned off.
Related pages
How to create a new agent
Step by step: project, agent node, instructions, model, tools, channel, testing and publishing.
Recipes: ready-made patterns
Six ready-wired starting projects: Telegram assistant, web app, mailbox, receptionist, agent with code, team of agents.
Governance and audit
Versions, rollback, activity log and tracing of every conversation: knowing what agents do.
Have an agent to take to production?Let’s talk.
We show you TeraBrain NX on your case: systems to connect, access and controls.