Skip to Content
Docs are being rebuilt — start at Introduction → How it works.
IntroductionCore conceptsAgent

Agent

An agent is one named AI employee. The simplest possible mental model: imagine hiring a colleague and writing them a one-page job description. The job description tells them who they are, what they do, what tools they have, and what documents to reference. The colleague — the agent — lives in your workspace and your team talks to it.

Every agent has:

  • A name that shows up in chat, channels, and audit logs (e.g. Hashim — Sales Coordinator).
  • A description — one sentence summarising what the agent does, shown in pickers and on the agent catalogue.
  • An instructions document — the prompt that shapes every reply. This is the heart of the agent.
  • A model — which underlying LLM powers it (you can pick OpenAI, Anthropic, Google, Z.AI, qwen-3.6 via api.taqat.qa, or any OpenAI-compatible endpoint).
  • Optional knowledge bases the agent can search.
  • Optional tools the agent can call (web search, code execution, sending email, Microsoft Teams, etc.).
  • Optional skills the agent can load on demand.
  • Optional channels that route messages to it.
  • Optional triggers that run the agent without a user.

The Agents page in the sidebar lists every agent in your workspace. Click one to open the agent builder; that’s where you configure everything above.

The two kinds of agents

  • Built-in specialists ship with the product — Atlas (technology Q&A), Mutarjim (Arabic ↔ English translation), Munaseq (Microsoft Teams coordinator), and so on. These appear automatically in your workspace and you can launch them on day one.
  • Custom agents you build yourself for your domain. The workflow is straightforward: Your first agent walks through it in 10 minutes.

You can also start from a built-in specialist, copy it, and edit — useful when a specialist is “almost right” for your team.

Agent vs Skill

  • An agent is one entity with one set of instructions. Use one agent per job.
  • A skill is a reusable behaviour pack an agent loads — e.g. “draft a weekly status update”. Use skills when two different agents share a common subroutine.

The SuperAgent

There’s one special agent you don’t see in the list: the SuperAgent (sometimes called the “generalist” or “TeamMate”). It’s the default specialist you talk to when you don’t pick a card from the chat empty state. It can dynamically discover and load other agents, tools, and skills from your workspace, so you don’t need to know who the right specialist is — just ask.

WorkspaceSkill