Canary Deployment Targeting Now Available for Telnyx AI Assistants

Telnyx AI Assistants now support end-user targeting in traffic distribution. You can route specific callers or destinations to a test version while keeping the rest of your traffic on the main version, then expand the rollout with percentage splits when you are ready.

What's new

• End-user target rules: Route calls by caller phone number, SIP URI, or destination. Use exact match, exclusion, or prefix operators to define which users receive a specific version.
• Percentage split with targeting: Combine targeting and percentage allocation in the same rule. Send 10% of matching traffic to a new version, keep 90% on main, then adjust the split as confidence grows.
• Ordered rule evaluation: Rules apply top to bottom, first match wins. Place specific test targets above broader rollout rules so internal numbers always reach the test version.
• Built-in fallback: Any unmatched percentage routes to the main version automatically. No configuration needed for safe rollback behavior.
• Extensible targeting framework: The rule engine supports more criteria in future releases. The current release focuses on end-user identifiers.

Why it matters

• Before this release, canary deployments used percentage splitting alone. You could send 10% of all traffic to a new version, but you could not guarantee your test number received it. Now you can target specific users directly, removing the randomness from validation.
• The typical workflow is: pin your test number to the new version, validate behavior, then expand to a percentage split for the same target group, then promote to main. This replaces the old pattern of separate dev and production environments.
• The rule-based approach scales with your testing maturity. Start with one test number, add a QA prefix, then layer on percentage splits. The same configuration surface handles every stage.

Example use cases

• A voice AI developer updates an assistant's instructions and routes their own phone number to the new version for 48 hours before enabling a 25% canary.
• A SaaS platform uses SIP URIs with a qa- prefix for internal testing. They configure a prefix rule so any URI starting with qa- always receives the latest version, while production endpoints stay on main.
• A contact center operator runs A/B tests between two assistant configurations. They split traffic 50/50 for a specific customer segment defined by originating number prefix, keeping all other traffic on the current version.
• A healthcare voice agent team validates a new tool integration. They target two compliance officer phone numbers to the new version, verify behavior under real call conditions, then gradually expand to the full patient-facing line.

Getting started

1. In Mission Control, navigate to AI > Assistants and select the assistant you want to configure.
2. Go to the assistant's version or deployment controls and open Traffic distribution.
3. Click Add rule. In the If section, choose End user target.
4. Select an operator: is one of, is not one of, or starts with.
5. Enter target values: phone numbers, SIP URIs, or prefixes. Separate multiple values with commas or new lines.
6. In the Serve section, choose Send all matched calls to one version and select your test version, or choose Split by percentage to assign weighted slots across versions. Percentages must add to less than 100; the remainder routes to the main version automatically.
7. Click Configure default to set fallback behavior for calls that do not match any target rule. By default, unmatched calls serve the main version. Use Reset to main to restore the default.
8. Drag rules to set priority order. Specific test targets should sit above broader rollout rules.
9. Click Save to apply the configuration.
10. Verify routing by making test calls. For inbound, call from a targeted number. For outbound, have the assistant call a targeted destination. For weighted rollouts, make enough calls to confirm traffic distributes according to your configured percentages.

Learn more in the Testing and Traffic Distribution documentation.