Skip to main content

What Are Agents?

Before diving into the architecture, it's worth stepping back and asking: what is an AI agent, and how is it different from a chatbot?

Chatbots vs. Agents​

You've probably chatted with a bot more times than you can count — asking for weather updates, tracking orders, or getting customer support. Modern chatbots, especially those powered by large language models, are pretty good at words. They generate human-like text based on what you ask. That's cool, but their capability mostly plateaus at conversation.

Agents are different. Sure, they can chat too, but their real strength is taking action. An agent:

  • Observes its environment — user requests, incoming emails, scheduled triggers, file changes
  • Decides what to do based on goals, tools available, and current context
  • Acts — calls APIs, writes files, sends emails, schedules follow-ups, updates databases
  • Reflects — checks whether the action worked, learns from results, updates its memory

Think of a chatbot as the smooth conversationalist at a party who can chat about anything but can't help you move a couch. An agent is the colleague who not only spots a problem but actually fixes it.

What an XpressAI Agent Looks Like​

On this platform, an agent is a containerized service with:

  • A role — defined by its system prompt (e.g., "you are a Sales Development Rep")
  • Tools — Python functions it can call (send email, search the web, create a Google Doc, run a shell command)
  • Memory — short-term (recent messages), mid-term (vector search via Vecto), long-term (markdown notes on disk)
  • A cognitive loop — the PFC orchestrator decides when to call tools, when to think harder, and when to respond
  • An identity — it shows up as a participant in conversations, can be @-mentioned, has its own email address, and can be assigned tasks

Agents run in their own Kubernetes pods, scale to zero when idle, and pick up work via RabbitMQ queues. See Agent Deployment Model for the gory details.

Why This Matters​

Most "AI products" today are wrappers around a chat API. They're useful, but limited — every interaction starts from zero, nothing happens in the background, and the user has to drive every step.

Agents flip that. You delegate a goal ("qualify these 50 leads, schedule meetings with the good ones"), and the agent works on it — taking actions, asking you when it needs input, reporting back when done. Multiple agents can collaborate in shared channels, each handling their specialty.

That shift — from chat partner to autonomous collaborator — is what makes agents qualitatively different.

Ready to Try It?​

Head to Create Your First Agent to deploy one from a template in under two minutes.