Follow the pre-deployment checklist, promote a draft to live, use traffic splitting for zero-downtime rollouts, and monitor the first real-world executions.
Pre-deployment checklist
Before deploying, verify the following: all test cases in your sandbox suite pass, the agent's production secrets and credentials are configured in the workspace's Secret Manager (not hardcoded), any external API integrations are connected with production-scoped tokens (not sandbox/test keys), the agent's error handler is configured to notify your team of failures, and you have set a realistic rate limit on the agent's trigger to prevent runaway execution. For scheduled agents, double-check the cron expression and timezone. For webhook agents, verify the webhook URL has been registered with the external system that will POST to it.
Promoting from draft to live
When you are ready to deploy, click Publish in the agent editor. Cotonity creates a versioned snapshot of the current agent graph and labels it with a version number (e.g., v1.0). The published version is immutable: subsequent edits create a new draft that does not affect the running agent until you publish again. After publishing, enable the agent's trigger by toggling the Trigger Active switch on the agent's overview page. From this moment, the agent will respond to incoming webhooks, scheduled events, or manual run requests. Keep the previous version in your run history as a reference in case you need to roll back.
Blue-green and canary deployments
For high-traffic agents where a breaking change could have significant impact, use Cotonity's traffic splitting feature to deploy a new version alongside the existing one. In the Deployments tab, create a new deployment pointing to the new version and set its traffic weight to a small percentage (e.g., 10%). The existing version handles the remaining traffic. Monitor both versions in the Metrics dashboard for error rates and response times. Once you are confident the new version is performing correctly, increase its traffic weight to 100% and retire the old version. Traffic splitting is available on Business and Enterprise plans.
Post-deployment monitoring
Immediately after deploying, watch the Live Runs panel for the first few real executions. Verify that trigger inputs are being received and parsed correctly, that all external API calls are completing successfully, and that the agent's outputs match expectations. Set up a monitoring alert in the agent's Settings > Alerts panel to notify you if the error rate exceeds a threshold (e.g., more than 5% of runs fail in a 15-minute window). Review the run logs at the end of the first day to catch any edge cases that your sandbox tests did not cover.