What Are Agents?
Understand how background agents work and what they can do
Background agents are AI assistants that run automatically without your direct involvement. They're like having a smart assistant who works around the clock.
The Basics
How Agents Differ from Chat
| Chat | Agent |
|---|---|
| You start the conversation | Agent runs automatically |
| Real-time interaction | Works in the background |
| One-time tasks | Recurring or triggered tasks |
| You're always present | Works while you're away |
What Agents Can Do
Agents can:
- Access your connected apps - Read emails, check calendars, post to Slack
- Process information - Summarize, analyze, transform data
- Take actions - Send messages, create documents, update records
- Make decisions - Use AI reasoning to handle variations
- Notify you - Send results via email or notification
What Agents Can't Do
Agents cannot:
- Access apps you haven't connected
- Make purchases or financial transactions (without explicit approval)
- Access your computer's local files
- Continue indefinitely (there's a timeout)
- Run without any trigger
How Agents Work
Trigger Activates
Something starts the agent:
- A scheduled time (e.g., 9 AM daily)
- A webhook call (e.g., from another app)
- A manual trigger (you click "Run Now")
Agent Receives Instructions
The agent loads its configuration:
- The instructions you wrote
- The AI model to use
- The tools and integrations it can access
Agent Executes
The AI works through your instructions:
- Reads from connected apps if needed
- Processes information
- Takes actions you've authorized
Agent Reports Results
When done, the agent:
- Logs what it did
- Sends notifications if configured
- Stores the result for you to review
Agent Components
Every agent has these parts:
1. Instructions (Prompt)
What you want the agent to do, written in plain language:
Check my Gmail inbox for emails from clients.
Summarize any urgent messages.
Post the summary to the #client-updates Slack channel.2. AI Model
The brain powering your agent. Different models for different needs:
- GPT-4o - Good for most tasks
- Claude 4 Sonnet - Great for writing and analysis
- GPT-4o Mini - Fast and efficient for simple tasks
3. Tools & Integrations
What the agent can access:
- Email - Gmail, Outlook
- Calendar - Google Calendar
- Messaging - Slack, Discord
- Productivity - Notion, Google Docs
- Development - GitHub, Linear
4. Trigger
When and how the agent runs:
- Schedule - Daily at 9 AM, every Monday, etc.
- Webhook - When called by another app
- Manual - When you click Run
5. Notifications
How you're informed:
- Email - Get results sent to your inbox
- In-app - See notifications in Billix
- None - Silent operation, check logs manually
Real-World Examples
Morning Briefing Agent
Instructions:
Every morning at 7 AM:
1. Check my Gmail for unread emails
2. Summarize any messages that need my attention
3. Check my calendar for today's meetings
4. Send me an email with the briefing
Trigger: Daily at 7:00 AM
Model: Claude 4 Sonnet
Tools: Gmail, Google Calendar
Notification: EmailPR Monitor Agent
Instructions:
When a new PR is opened in our GitHub repo:
1. Read the PR description and changes
2. Post a summary to #engineering Slack
3. Tag the team lead if it's a large PR
Trigger: Webhook from GitHub
Model: GPT-4o
Tools: GitHub, SlackWeekly Report Agent
Instructions:
Every Friday at 5 PM:
1. Gather data from our Notion workspace
2. Calculate key metrics for the week
3. Create a formatted report
4. Post to the #reports channel
Trigger: Weekly, Fridays at 5 PM
Model: GPT-4o
Tools: Notion, SlackAgent Execution
Timeouts
Agents have a maximum execution time of 5 minutes. For longer tasks:
- Break into smaller agents
- Use multiple scheduled runs
- Process data incrementally
Retries
If an agent fails (network error, API issue), Billix automatically retries:
- First retry: After 1 minute
- Second retry: After 5 minutes
- Third retry: After 15 minutes
- After 3 failures: Agent stops, you're notified
Costs
Agent runs use credits based on:
- The AI model used
- The length of processing
- Number of tool calls
Most simple agents cost 5-20 credits per run.
Security
What Agents Can Access
Agents can only use:
- Integrations you've specifically connected
- Tools you've explicitly enabled for that agent
- Data within those integrations' permissions
What's Protected
Agents cannot:
- Access integrations you haven't connected
- See data outside their granted permissions
- Share data with external services (unless you configure it)
- Run without proper authentication
Common Questions
Next Steps
Ready to build? Learn how to create your first agent.
Was this page helpful? Let us know!
Report an issue