hi

Thanks for reaching out, Arun. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

i want to self host my live kit agents but i want to use livekit cloud for sfu server is jt pissivleb abd what is the cost

Thanks — I’m glad you asked. I’ll answer both parts: feasibility and cost. Acknowledgement • You asked whether you can self‑host your LiveKit agents while using LiveKit Cloud as the SFU, and what that will cost. Is this possible? • Yes — you can self‑host your agents (your AI/LLM orchestration and application logic) and connect them to LiveKit Cloud for media routing. LiveKit Cloud uses the same SFU, APIs, and SDKs as the open‑source server, so agents can connect to Cloud rooms without changing your code. You can even split traffic between self‑hosted SFU and Cloud if you want. How billing works (what you will pay) • When using LiveKit Cloud for the SFU, you are billed for media transport usage, not for where your agents run. Main billing components are: ◦ Connection time (participant minutes): billed per minute per participant (each agent counts as a participant while connected). ◦ Downstream bandwidth: billed per GB for media sent from Cloud to participants. ◦ Upstream bandwidth: currently free. • If you use LiveKit’s Inference features (LLM/STT/TTS via LiveKit), that is billed separately. If you self‑host and call external AI APIs yourself, you only pay LiveKit for connection time and downstream bandwidth. Representative pricing (example rates) • Connection time: $0.0005 per minute per participant (rate scales with volume). • Downstream bandwidth: $0.12 per GB (volume discounts may apply). • Upstream bandwidth: free. What this means for your agents • Each agent connected to a Cloud room is billed as a participant. If an agent stays continuously connected, its minutes will accumulate and be billed. • If your agent performs STT/TTS via LiveKit Inference, expect additional inference charges. If you’d like an estimated monthly cost • Tell me: a. How many agents will be connected concurrently (average and peak). b. Average session length per agent (minutes per connection). c. Expected downstream bandwidth per session (approx GB per session) or whether sessions are audio‑only vs audio+video. d. Whether you’ll use LiveKit Inference or self‑host calls to external LLM/STT/TTS providers. With those numbers I can give a simple estimate (monthly cost for LiveKit Cloud SFU + bandwidth and note any additional inference cost if used). Would you like me to run that estimate for you? Sources: Announcing LiveKit Cloud | The end of Participant Minute pricing | Introduction | LiveKit Documentation | Google AI and LiveKit | LiveKit Documentation | Towards a future-aligned pricing model