Getting Started
Agent Orchestrator lets you build automated workflows using simple HTTP APIs, webhooks, databases, and more. Create agents that run on schedules or via API calls.
Core Concepts
Agents
An agent is a workflow made up of one or more steps. Each step uses a tool (HTTP, webhook, database, etc.) to perform an action. Steps run sequentially and can access data from previous steps.
Tools
Tools are the building blocks of agents. Each tool performs a specific action:
- HTTP - Make API calls to external services
- Webhook - Send data to webhook endpoints
- Database - Query PostgreSQL databases
- SendGrid - Send emails via SendGrid
- Twilio - Send SMS messages via Twilio
- Transform - Manipulate and transform data
- Conditional - Branch logic based on conditions
- Delay - Wait for a specified time
- LLM - Call OpenAI or Anthropic models for AI tasks
Context Variables
Access data from previous steps using template syntax:
{{step0.data}} - Output from first step
{{step1.status}} - HTTP status from second step
{{_trigger.body}} - Data from API trigger
{{_workspace.name}} - Workspace information
Tool Configuration
HTTP Tool
{
"url": "https://api.example.com/data",
"method": "GET",
"headers": {
"Authorization": "Bearer YOUR_TOKEN"
},
"body": {
"key": "value"
}
}
Webhook Tool
{
"url": "https://hooks.slack.com/services/YOUR/WEBHOOK",
"method": "POST",
"body": {
"text": "Message from agent: {{step0.data}}"
}
}
Database Tool
{
"connection_string": "postgresql://user:pass@host:5432/db",
"query": "SELECT * FROM users WHERE id = $1",
"params": ["{{_trigger.userId}}"]
}
SendGrid Tool
{
"from": "you@example.com",
"to": "recipient@example.com",
"subject": "Alert",
"text": "Status: {{step0.status}}",
"html": "<h1>Status: {{step0.status}}</h1>",
"api_key": null
}
Set api_key to null to use workspace SendGrid API key from Settings. Get your key from SendGrid.
Twilio Tool
{
"from": "+1234567890",
"to": "+1234567890",
"body": "Alert: {{step0.status}}",
"account_sid": null,
"auth_token": null
}
Set credentials to null to use workspace Twilio credentials from Settings. Get credentials from Twilio Console.
LLM Tool
{
"provider": "openai",
"prompt": "Summarize this: {{step0.data}}",
"model": "gpt-4o-mini",
"temperature": 0.7,
"max_tokens": 1000,
"api_key": null
}
Supported providers: openai (GPT models) or anthropic (Claude models). Set api_key to null to use workspace LLM API key from Settings.
Running Agents
Via Webhook (HTTP)
POST /v1/runs
Authorization: Bearer YOUR_API_KEY
Content-Type: application/json
{
"agent_id": "agent_abc123",
"project_id": "proj_xyz789",
"input": {
"userId": "123",
"action": "process"
}
}
Via Email
Send email to agent-{agentId}@yourdomain.com to trigger the agent. The email content (from, subject, body) will be passed as input.
Setup: Configure SendGrid Inbound Parse to forward emails to https://your-domain.com/v1/inbound/email
Via SMS
Send SMS to your Twilio number to trigger the agent. The message content will be passed as input.
Setup:
- Configure Twilio webhook:
https://your-domain.com/v1/inbound/sms - Set Railway env var:
AGENT_PHONE_MAPPINGS=agentId:+1234567890
Via Calendar
Trigger agent when Google Calendar events change.
Setup:
- Set up Google Calendar API push notifications to
https://your-domain.com/v1/inbound/calendar - Set Railway env var:
AGENT_CALENDAR_MAPPINGS=agentId:channelId - Use Google Calendar API to watch calendar:
POST https://www.googleapis.com/calendar/v3/calendars/primary/events/watch
Via Schedule
Set up recurring runs using cron expressions:
0 9 * * *- Daily at 9 AM*/15 * * * *- Every 15 minutes0 0 * * 0- Weekly on Sunday at midnight
Usage Limits
| Plan | Runs/Month | Steps/Month | HTTP Calls | Webhooks | Execution Hours |
|---|---|---|---|---|---|
| Free | 100 | 1,000 | 1,000 | 100 | 1 |
| Pro | 10,000 | 100,000 | 100,000 | 10,000 | 100 |
| Enterprise | Unlimited | Unlimited | Unlimited | Unlimited | Unlimited |
Best Practices
- Use templates to access data from previous steps
- Store sensitive credentials in workspace settings when possible
- Set appropriate timeout values for long-running operations
- Use retry policies for unreliable external APIs
- Test agents with small datasets before scaling up
- Monitor usage to stay within plan limits
Support
Need help? Check out these resources:
- Browse templates for working examples
- View run logs to debug failures
- Check usage stats in your dashboard