Skip to main content

​
Core Tools

Core tools give agents the ability to manage your Celeria workspace, turning them into assistants for building agentic workflows. Agents with core tools can:
  • Execute tasks - Trigger other tasks to spawn new sessions with different agents
  • Create/Modify resources - Add or modify agents, actions, and tasks
  • List resources - Query available models, integrations, agents, actions, and tasks
This makes agents with core tools useful for orchestrating complex workflows and building systems that manage themselves.

​
The Default Agent

When you link your first inference provider, Celeria creates a default agent named “Celeria” with:
  • Core tools enabled
  • Your provider’s latest model
  • Organization-wide default status
This agent is ready to help you build through conversation.

​
Building Through Conversation

You can build everything by talking to your default agent. You also have full control to create and modify anything directly through the UI. You can create resources through conversation:
“Create an agent called ‘Weather Reporter’ with access to web fetch and my Slack integration.”
Or query your workspace:
“Show me all my agents” “What tasks are running?”
Here’s an example of building a daily weather report automation:

​
Create an Agent

Plain English Prompt:
“Create an agent called ‘Weather Reporter’ with access to web fetch and my Slack integration.”
Equivalent Configuration: 
{
  "name": "Weather Reporter",
  "systemPrompt": "You are a helpful weather reporter that fetches weather data and presents it in a friendly, concise format.",
  "model": "claude-sonnet-4",
  "celeriaToolsEnabled": false,
  "integrations": ["slack", "fetch"]
}

​
Build an Action

Plain English Prompt:
“Create an action called ‘Weather Report’ with a variable for location. The action should fetch weather data from weather.com for the given location, summarize it into a friendly weather report format, and send it to #general in Slack.”
Equivalent Configuration: 
{
  "name": "Weather Report",
  "variables": {
    "location": "string"
  },
  "messages": [
    {
      "role": "user",
      "content": "Fetch the current weather for {{location}} from weather.com, summarize it in a friendly format with temperature, conditions, and any relevant alerts, then post it to the #general channel in Slack."
    }
  ]
}
Actions are reusable prompt templates. The {{location}} variable can be filled in when the action is used.

​
Set Up a Task

Plain English Prompt:
“Create a task called ‘Daily Weather’ that runs every day at 9 AM EST using the Weather Reporter agent with the Weather Report action. Set the location to South Bend, IN.”
Equivalent Configuration: 
{
  "name": "Daily Weather",
  "agent": "Weather Reporter",
  "triggers": [
    {
      "type": "cron",
      "config": {
        "schedule": "0 9 * * *",
        "timezone": "America/New_York"
      }
    }
  ],
  "taskActions": [
    {
      "action": "Weather Report",
      "order": 1
    }
  ],
  "variables": {
    "location": "South Bend, IN"
  },
  "isEnabled": true
}
The task combines the schedule trigger, agent, and action. Each time it runs at 9 AM EST, it creates a new session where the Weather Reporter agent executes the Weather Report action with location set to “South Bend, IN”.

​
Best Practices

Provide complete context. Include the agent name, model, and integrations in your request: “Create an agent called ‘Code Reviewer’ using Claude Sonnet 4.5 with access to my GitHub integration.” Start small and compose. Build simple, focused agents and actions that do one thing well. Combine them through tasks and orchestration to create powerful compound workflows that scale. Iterate through observation. Run your agent in a session before automating. Watch what works and what doesn’t. Refine your system prompts and action templates based on actual behavior, not assumptions. Grant capabilities intentionally. Core tools let agents modify your workspace. Only enable them when you need orchestration or self-modification. Keep task-execution agents focused on their specific domain.

​
Next Steps

Explore specific concepts:
  • Agents - Configure AI with custom instructions, models, and capabilities
  • 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 - Set up autonomous agentic workflows