Skip to main content

Deploying your agent

Once you've built and tested your agent, you're ready to deploy it to handle real customer conversations.

Production vs. testing​

  • Playground: Browser-based voice conversations for internal testing and development. Select an environment to test with different configurations.
  • Public link: Shareable link for stakeholder feedback and beta testing with limited users.

Production: Phone system integration and SDK​

  • Phone connectors: Route inbound phone calls from your telephony system (Genesys, Twilio, or SIP) to your agent. See phone system integration options below.
  • Agent SDK: Embed real-time voice conversations directly in your web or mobile app using the @stellar-ai/agent-sdk package.
important

Building and testing your agent doesn't automatically deploy it to customers. Production deployment requires setting up a phone connector or integrating the SDK.

Phone system integration options​

Stellar integrates with major phone systems to route calls to your AI agents. Phone integrations are managed via the Phone connectors page in Studio.

Supported platforms​

  • Genesys Cloud: Audio connector integration for Genesys environments
  • Twilio: Direct integration for Twilio-based phone systems
  • SIP providers: Support for other telephony systems via SIP protocol

Connectors​

A connector maps an external identifier (phone number, SIP URI, etc.) to a specific agent, environment, and version. You create and manage connectors on the Phone connectors page in Studio.

Each connector has:

  • Type: Genesys, Twilio, or SIP
  • Agent: Which agent handles calls through this connector
  • Environment: Which environment (Development, Staging, Production) to use. This determines which auth connector configs, secret values, and variable values are resolved for calls.
  • Version mode: Published (latest published version), Draft, or Pinned to a specific agent version.

How integration works​

  1. Customer calls your business phone number
  2. Your phone system routes the call to Stellar via the connector
  3. Your AI agent answers using the connector's environment and version settings
  4. If needed, the call transfers back to a human agent
  5. Full conversation with your Stellar agent is recorded and available for review

Deployment process​

Step 1: Prepare your agent​

Before deploying, ensure:

Testing is complete

  • Extensive playground testing across all environments
  • Public link testing with stakeholders
  • All major scenarios verified
  • Edge cases handled appropriately

Configuration is finalized

  • Personality and voice are set
  • Knowledge base is comprehensive
  • Dialogue flows cover all scenarios
  • Actions are tested and working
  • Guardrails are configured
  • Evaluation rules are in place
  • Per-environment secrets, variables, and auth connectors are configured

Step 2: Create a connector​

  1. Navigate to Phone connectors in Studio
  2. Click Add connector
  3. Select the connector type (Genesys, Twilio, or SIP)
  4. Choose your agent
  5. Select the environment (Development, Staging, or Production)
  6. Choose the version mode (Published, Draft, or Pinned)
  7. Save the connector

The connector generates an identifier used to route calls from your phone system to Stellar. See the platform-specific setup guides for details:

Step 3: Configure your phone system​

Use the connector's identifier to configure your phone system to route calls to Stellar. The exact steps depend on your platform — see the guides linked above.

Step 4: Test the integration​

Before going live:

  • Make test calls through the phone system
  • Verify the AI agent answers and handles conversations
  • Test handover flows
  • Verify audio quality and reliability
  • Check that the correct environment config is being used

Step 5: Go live​

Once testing is complete:

Gradual rollout (recommended)

  • Start with a small percentage of calls
  • Monitor performance closely
  • Increase volume as confidence grows
  • Full rollout once proven

Step 6: Monitor and optimize​

After deployment:

  • Review conversations daily
  • Check evaluation scores
  • Watch for unexpected issues
  • Track handover rates
  • Update agent based on real-world performance

Deployment considerations​

Call volume planning​

Estimate your call volume:

  • Current average calls per day
  • Peak times and days
  • Seasonal variations
  • Growth expectations

Stellar can scale to handle your volume, but it helps to plan ahead.

Human agent backup​

Plan for human agent availability:

  • Handover capacity during peak times
  • Escalation paths for complex issues
  • Coverage for situations AI can't handle
  • Training for human agents on takeover process

Business hours​

Decide when the AI agent should be active:

  • 24/7 availability
  • Business hours only
  • Different hours for different call types
  • Holiday and special event handling

Routing strategy​

Determine how calls should be routed:

  • All calls to AI first, then escalate if needed
  • Certain call types to AI, others to humans
  • Customer choice (IVR menu)
  • Time-based routing (AI after hours, humans during business hours)

Testing in production​

Even after deployment, continue testing:

Shadow testing​

  • Run AI in parallel with humans initially
  • Compare outcomes and quality
  • Identify areas for improvement
  • Build confidence before full deployment

A/B testing​

  • Route some calls to AI, others to humans
  • Compare customer satisfaction
  • Measure efficiency gains
  • Validate business impact

Continuous validation​

  • Regularly test your phone number
  • Verify the experience customers have
  • Check for any degradation
  • Ensure quality remains high

Rollback plan​

Have a plan to revert if needed:

If issues arise:

  1. Route calls back to humans immediately
  2. Investigate and fix the problem
  3. Test the fix thoroughly
  4. Re-deploy when ready

Common reasons for rollback:

  • Unexpected technical issues
  • Agent performing worse than expected
  • Phone system integration problems
  • Business requirement changes

Multi-agent deployments​

If you have multiple agents:

Agent per purpose​

  • Different agents for different departments
  • Billing agent, technical support agent, sales agent
  • Route based on caller selection or automated detection

Agent per channel​

  • Different agents for different phone numbers
  • Main line vs. support line vs. sales line
  • Each optimized for its specific purpose

Agent versions and environments​

Use connectors to manage different deployment scenarios:

  • Production connector: Uses the Production environment with the Published agent version for live customer calls.
  • Staging connector: Uses the Staging environment with a pinned version for pre-production validation.
  • Development connector: Uses the Development environment with the Draft version for experimentation.

Deployment best practices​

Start small​

  • Deploy to limited call volume first
  • Specific call types or times
  • Lower-risk scenarios
  • Build confidence gradually

Monitor intensively at launch​

  • Watch every conversation initially
  • Use listen & intervene liberally
  • Gather feedback quickly
  • Fix issues immediately

Communicate with customers​

Consider informing customers:

  • "You'll be speaking with our AI assistant"
  • Set appropriate expectations
  • Offer human option if preferred
  • Be transparent about capabilities

Train your team​

Ensure human agents understand:

  • When they'll receive handovers
  • How to use listen & intervene
  • What to expect from AI-assisted handoffs
  • How to provide feedback on AI performance

Plan for success​

Anticipate positive outcomes:

  • Reduced wait times
  • Increased customer satisfaction
  • Higher efficiency
  • 24/7 availability
  • Data and insights from conversations

Post-deployment​

After successful deployment:

Measure impact​

Track key metrics:

  • Call volume handled by AI
  • Handover rates
  • Customer satisfaction
  • Average handle time
  • Cost savings
  • Quality scores

Continuous improvement​

Use production data to:

  • Identify knowledge gaps
  • Refine dialogue flows
  • Update guardrails
  • Add new capabilities
  • Optimize performance

Scale and expand​

Once successful:

  • Increase call volume to AI
  • Deploy additional agents
  • Expand to new call types
  • Add more integrations

Getting help​

For deployment support:

  • Contact the Stellar team for:

    • Integration setup
    • Technical configuration
    • Troubleshooting
    • Best practices guidance
    • Scaling advice

  • Refer to specific integration guides:

Next steps​

Ready to deploy? Here's what to do:

  1. Finalize your agent configuration
  2. Complete thorough testing
  3. Review these deployment docs
  4. Contact the Stellar team to begin
  5. Follow the deployment process
  6. Monitor and optimize post-launch