https://livekit.io logo
Join Slack
Powered by
Is the agent in sandbox, an exact replica of the p...
# ask-ai
s
Is the agent in sandbox, an exact replica of the production agent once created? What’s the ci/cd process from sandbox to production?
t
Thanks for reaching out, Saeed. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
You're asking whether a sandbox/preview agent is an exact replica of a production agent, and what the CI/CD path is from sandbox to production — here's the concise answer. • Sandbox/preview agent behavior ◦ Every agent created with the Agent Builder is a real, fully functional LiveKit Agent running on LiveKit Cloud (same voice pipeline, phone number integration, observability, and client SDKs). ◦ The preview session used while building is running on that real backend but is intended for development: it uses your project's Inference credits and preview sessions do not appear in Agent Observability or usage metrics. • Is the sandbox an exact replica of production? ◦ Functionally yes: the agent runs the same production-grade runtime and features. ◦ Practically there is a distinction: preview sessions are isolated for development (they don’t appear in observability/usage). Full production behavior (visible sessions/metrics, routing for live users) occurs after deployment. • CI/CD / deployment workflow a. Build and test in the Agent Builder using preview mode. b. To make an agent live, click “Deploy agent” in the Agent Builder — this makes it a full production instance. c. Optionally download the agent code (ZIP) from Agent Builder to get a complete Python project. d. Move to code-based deployment and use the LiveKit CLI (
lk agent deploy
) so you can integrate the agent into your version control and standard CI/CD pipelines. e. For code-deployed agents on paid plans you can update with 
lk agent deploy
and revert with 
lk agent rollback
(rolling updates / rollback support). If you want, I can: • Walk through exporting the agent code and a sample GitHub Actions or other CI/CD pipeline for deploys, or • Explain differences you should watch for when promoting changes (e.g., observability, credentials, phone number config). Which would you prefer? Sources: LiveKit Blog | https://blog.livekit.io/announcing-agent-builder-a-pathway-to-code/#/portal| | Announcing Agent Builder: A Pathway to Code | LiveKit Blog | Agent Builder | LiveKit docs
s
So what exact credits does the sandbox env use
t
Preview (sandbox) sessions use your project's LiveKit Inference credits. They do not appear in Agent Observability or count toward usage metrics, but they still consume Inference credits from your project.
s
since they arent in agent observability how can i see the traces and hear the audio recordings
t
You're asking how to view traces and audio recordings for sandbox/preview sessions. Short answer: • Preview/sandbox sessions do not appear in Agent Observability, so you cannot view their traces or synchronized audio recordings. • If you need traces, transcripts, or audio playback you must deploy the agent to production. Once deployed, sessions appear in Agent Insights (project sessions dashboard) where you can view turn-by-turn traces, synchronized audio, transcripts, logs, and download session data. Would you like steps to deploy from the Agent Builder or help exporting the agent code so you can add CI/CD and preserve observability? Sources: Agent Builder | LiveKit docs | Streamline Troubleshooting with Agent Observability | Agent insights in LiveKit Cloud | LiveKit docs | Quotas and limits | LiveKit docs
4 Views
Open in Slack
PreviousNext
Powered by