Skip to main content
An agent is a reusable AI configuration that lives in your workspace. Instead of repeating the same context in every chat, you define it once — and the agent remembers it every time.

Agent Tasks

Automate your agents with tasks

Triggers

Schedule and event-based automation

​
What makes an agent

Every agent is built from a few key components:
ComponentWhat it does
InstructionsThe system prompt — defines the agent’s role, tone, and behavior
ModelWhich AI model powers this agent (GPT, Claude, Gemini…)
KnowledgeFiles you upload as context (docs, guidelines, reference material)
ToolsCapabilities you enable — web search, MCP integrations, connected apps
TasksAutomations attached to this agent — when and how it runs automatically

​
Agents vs. plain chats

ChatAgent
ContextYou set it each timeDefined once, always applied
ReuseStart fresh every timeClick to use across your workspace
AutomationNot possibleTasks let it run automatically
TeamPersonalShared with your workspace
Use a plain chat when you need something quick and one-off. Use an agent when you want consistent behavior across many conversations — or when you want to automate the agent with a task.

​
Creating an agent

1

Open the Agents panel

In your workspace sidebar, click Agents (or the agent icon).
2

Click New Agent

Give it a name that describes its purpose — “Research Assistant”, “Code Reviewer”, “Daily Digest”.
3

Write its instructions

The instructions become the agent’s system prompt. Be specific:
You are a senior code reviewer. Focus on security, performance, and maintainability.
Always suggest concrete improvements with code examples.
Use our team's style guide (attached in knowledge).
4

Choose a model

Pick the AI model that fits the task. Claude is great for writing and reasoning; GPT-4o for general tasks; Gemini for research.
5

Add knowledge (optional)

Upload files that the agent should always have access to — style guides, API docs, onboarding materials.
6

Enable tools (optional)

Turn on web search, or connect MCP integrations and apps the agent needs.
7

Add tasks (optional)

Go to the Tasks tab to automate the agent — run it on a schedule, on demand, or when an event happens.

​
Agents and tasks

An agent becomes an automation when you attach a task to it. 
Agent (the who + what)
  └── Task (when it runs + where output goes)
        ├── Trigger: every morning at 9am
        └── Output: post to #team-digest channel
One agent can have multiple tasks — each for a different automation: 
Agent: "Research Assistant"
  ├── Task: "Daily digest"     → runs 9am daily → posts to channel
  ├── Task: "Weekly summary"   → runs every Monday → creates new chat
  └── Task: "On demand"        → run when you press a button
The agent’s instructions and tools stay the same. Each task can add extra instructions on top for that specific run.

Learn about tasks

How to set up automations on your agents

​
Sharing agents

Agents in a shared workspace are visible to all workspace members by default.
  • Team members can use any public agent to start conversations
  • Admins and owners can edit agents created by anyone
  • Private agents are only visible to you
Create a shared “Code Reviewer” or “Meeting Notes” agent once, and your whole team benefits from the same consistent configuration.

​
Example agents

Reviews pull requests, catches security issues, and suggests improvements following your team’s standards.Model: Claude 3.5 Sonnet (strong reasoning) Knowledge: Team style guide, security checklist Tools: None needed for review Tasks: Manual — run when you want a review
Gathers updates from your sources every morning and posts a structured summary to your team channel.Model: GPT-4o Tools: Web search Tasks: Schedule — daily at 8am → post to #digest
Compiles metrics, highlights trends, and posts a formatted report every Monday.Model: GPT-4o Tools: Connected data apps Tasks: Schedule — every Monday 9am → post to #reports
Handles first-line support with deep knowledge of your product and policies.Model: Claude 3.5 Sonnet Knowledge: Product docs, FAQ, refund policy Tools: None — response only Tasks: Manual — run when needed
Deep-dives into any topic, synthesizes sources, and produces structured research briefs.Model: Claude Opus or GPT-4o Tools: Web search Tasks: Manual — run on demand