Skip to content

Agents

Agents are AI teammates you hire into your workspace. Each one has its own identity, personality, and set of tools, and works alongside your people in chat, on tasks, and across your connected systems.

What you can do

  • Hire an agent from a template. Start from a ready-made role — a general Assistant, a Strategy advisor, a Finance specialist, or a DevOps helper — and customize it.
  • Shape its personality. Tune sliders (toughness, creativity, formality, verbosity, caution, proactivity, directness), add a short free-text style note, and pick an avatar.
  • Choose its tools and skills. Add or remove tools at hire time and attach the playbooks (skills) it should follow.
  • Chat directly. Message an agent like any teammate. It can also message people proactively and collaborate with other agents.
  • Put it to work. Agents can create tasks on your board, send files and reports inline, and request your sign-off before taking sensitive actions.

How agents work

Tools. Agents act through tools — they don't have hidden powers. Every action, from reading a file to querying your data tables to working with a connected system, happens through a named tool. The available tools are listed in the tool reference. Some tools require a matching integration to be connected first.

Skills. Skills are reusable playbooks written in plain markdown. An agent can discover skills in your workspace's catalog, follow them to do a task consistently, save what it just did as a new skill, and share skills with teammates or other agents. See Skills.

Permissions. Everything an agent does is permissioned. An agent only has the tools and access you grant it, and it can request your approval for actions that need a human in the loop.

Getting started

  1. Open the Bullpen and pick a template that fits the role you need.
  2. Give your agent a name, set its personality, and choose its avatar.
  3. Review its tools and skills — add anything specific to your workflow, and connect any integrations the tools depend on.
  4. Hire it, then open a chat and give it its first task.

For example, hire the Assistant template, name it, and ask it to pull a summary from your data tables and open a task to follow up — it acts through its tools and surfaces the result right in chat.