Skip to main content

​
What Are Agents?

Agents are reusable configurations that define how an AI should behave and what it can do. Each agent combines a system prompt, model selection, and integrations into a single configuration you can use across multiple sessions and tasks. Think of agents as specialized assistants. You create them once with specific instructions and capabilities, then deploy them wherever needed.

​
Creating an Agent

You can create agents through conversation with your default agent, or directly in the UI by clicking “New Agent” from the Agents page.

​
Configuration

Name (required) - A descriptive identifier for your agent (2-100 characters) System Prompt (required) - Instructions that define your agent’s role, expertise, and behavior (10-5000 characters). This is the foundation of your agent’s personality and capabilities. Model (required) - The AI model your agent uses. Available models depend on which inference providers you’ve connected (Anthropic, OpenAI, etc.). Each model shows its provider and cost per million tokens. Integrations (optional) - External services and tools your agent can access. Choose from OAuth integrations like Gmail, Slack, and GitHub, or add custom integrations using MCP (Model Context Protocol) servers. Celeria Tools (optional) - Enable this to give your agent access to core platform tools. Agents with core tools can create and modify other agents, actions, and tasks, making them useful for orchestration and workspace management.

​
System Prompts

The system prompt is where you define what your agent does and how it behaves. Good system prompts are specific about:
  • Role and expertise - What domain knowledge does this agent have?
  • Task scope - What should the agent focus on or avoid?
  • Tone and style - How should the agent communicate?
  • Constraints - What rules or limitations should the agent follow?

​
Example: Customer Support Agent

You are a customer support specialist for an e-commerce platform.
Your role is to help customers with order status, returns, and general
product questions.

When responding:
- Be friendly and empathetic
- Provide clear, actionable solutions
- Escalate to human support for refunds over $100 or account
  security issues
- Always verify order details before making changes

You have access to the order management system and can look up
orders, update shipping addresses, and initiate returns.

​
Integrations

Give your agents access to any of your connected integrations. Agents can use whatever integrations you’ve set up in your workspace - OAuth services like Gmail, Slack, and GitHub, or custom integrations using MCP (Model Context Protocol) servers. See Integrations for details on connecting external services.

​
Celeria Tools

Celeria Tools are special - they’re not external integrations, but core platform capabilities that let agents manage your workspace. When enabled, agents get access to these tools:
  • list_agents - Query all agents in the workspace
  • get_agent - Get details about a specific agent
  • create_agent - Create new agents
  • edit_agent - Modify existing agents
  • list_actions - Query all actions in the workspace
  • get_action - Get details about a specific action
  • create_action - Create new actions
  • edit_action - Modify existing actions
  • list_tasks - Query all tasks in the workspace
  • get_task - Get details about a specific task
  • create_task - Create new tasks
  • edit_task - Modify existing tasks
  • execute_task - Trigger a task to spawn a new session
  • list_available_models - See what models are available
  • list_integrations - See what integrations are connected
This makes agents with core tools powerful for orchestration and building self-modifying systems. Only enable core tools when you need this capability, as it grants significant control over your workspace.

​
Default Agent

Your organization has one default agent that’s used when you start a new session without specifying which agent to use. The default agent is created automatically when you connect your first inference provider with:
  • Core tools enabled
  • Your provider’s latest model
  • Organization-wide default status
You can change which agent is the default, and you can update the default agent’s configuration (system prompt, model, integrations) just like any other agent.

​
Using Agents

Once created, agents can be used in two ways:

​
In Sessions

Start a new session and select your agent. The session will use that agent’s configuration for all interactions. You can observe what the agent does, see tool calls, and interact in real-time.

​
In Tasks

Assign an agent to a task. When the task executes (via cron schedule, webhook, or manual trigger), it creates a session using that agent to perform the work.

​
Agent Access

Agents are scoped to your organization. All members can view and use any agent, but only admins can create, edit, or delete them.

​
Next Steps

  • Sessions - Observe and interact with agents in real-time
  • Integrations - Connect to external services and data sources
  • Actions - Create reusable prompts for your agents to execute
  • Tasks - Automate agent workflows with triggers and schedules