Agents

An agent is the AI you've connected to withHuman. It could be helping you write code, handling customer support, or doing something else entirely. You give it a name so you can recognise it when a request comes in.

Agent instances

You can run the same agent in more than one place. Each connected copy is an agent instance. That might be the one on your laptop, the one on a server, or the one running an automated job.

Imagine your support agent runs on two servers. That's one agent with two instances. Both do the same job, but you can tell them apart.

Or imagine an AI SRE agent that spins up a new copy in the cloud for each incident. You'd have one agent, and each new copy would be an instance of it.

Those are service agents: the organization runs them, and their instances belong to nobody in particular. A coding assistant on someone's laptop is different. It's a personal agent: each instance belongs to the member who connected it and acts on their behalf, so its requests say who it was working for, and the instance goes when they leave. An agent is one or the other from the moment it's created.

flowchart TD
    agent["Support agent"] --> first["Instance on server 1"]
    agent --> second["Instance on server 2"]

Agents and rules

Grouping instances under an agent allows you to organize them easily. An approval pipeline can check which agent sent a request, so you can define rules once for a particular agent, then just add new instances as you need them and they automatically have the same rules applied.