Hi

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

Can I combine self hosted Livekit agent while using Livekit cloud for telephony and voice transport? If yes, how?

Thanks — great question. You can run a self‑hosted LiveKit agent while using LiveKit Cloud for telephony/voice transport. Below is a concise explanation and the steps to make it work. What this looks like • LiveKit Cloud handles SIP/telephony and creates the room with the caller as a SIP participant. • Your self‑hosted agent runs on your infrastructure but connects to the LiveKit Cloud project and joins the cloud room to interact with the caller. • Voice/media flows through LiveKit Cloud’s transport; the agent simply joins the room like any other participant. How to set it up 1. Configure your agent to use your LiveKit Cloud URL (wss://<yourproject>.livekit.cloud) and the Cloud project API key/secret. 2. Deploy your agent server on your infrastructure and ensure it has outbound network access to LiveKit Cloud. 3. Register the agent with LiveKit Cloud (use the Agents framework/registration flow) so dispatch rules can target it. 4. Create dispatch rules in LiveKit Cloud to route SIP calls to your agent when calls arrive. 5. When a SIP call arrives, LiveKit Cloud creates the room and dispatches your registered agent; the agent connects to the room and handles audio in real time. Requirements & notes • The agent must be able to reach LiveKit Cloud over websocket/TCP. • Use the standard LiveKit SDK/APIs for the agent — the same APIs work whether the agent is self‑hosted or cloud‑connected. • This setup lets you keep agent logic and sensitive processing on your infrastructure while leveraging LiveKit Cloud for telephony and global voice transport. If you’d like, tell me: • which Agents SDK or sample you’re using (language), and • your LiveKit Cloud project hostname — I can give the exact connection snippet and the key places to set your project API key/secret and dispatch rules. Sources: SIP overview | LiveKit docs | LiveKit Docs | Agents telephony integration | LiveKit docs | LiveKit Cloud | LiveKit Docs | OpenAI and LiveKit partner to turn Advanced Voice into an API

what all LiveKit cost components will apply (mandatory and optional). Telephony will be exotel, STT, TTS, LLM will be through plugins and not relevant as agents will be self hosted

Thanks — I’ll outline which LiveKit cost components will apply to your setup (Exotel for telephony, STT/TTS/LLM via plugins). Mandatory LiveKit charges • WebRTC participant minutes — charged per minute per participant who connects to LiveKit rooms (humans and agents) • Downstream bandwidth (egress) — charged per GB for data sent from LiveKit to participants Potential LiveKit charges depending on choices • Agent session minutes — charged if you host agents on LiveKit Cloud (per‑minute rate for agent sessions). If your agent runs fully on your infra but connects to the cloud only for transport, you still incur participant minutes for that agent. • Egress for recordings/streaming — if you record sessions or stream out to external platforms, egress and recording storage/processing fees may apply • Observability/recording add‑ons — optional features (recordings, events, observability) have usage fees if enabled beyond free tiers What you will NOT be charged by LiveKit (since you’re using external services) • Telephony call minutes, numbers, and trunking — billed by Exotel, not LiveKit • STT/TTS/LLM inference — billed by whichever plugin/provider you use; LiveKit won’t charge for those if you don’t use LiveKit’s inference services If you want, tell me: • expected concurrent participants and average call length, and • whether agents will run on LiveKit Cloud or entirely on your infra I can then estimate monthly WebRTC minutes and egress to help with cost planning. Sources: LiveKit Pricing | Towards a future-aligned pricing model | Deploy and scale agents on LiveKit Cloud | Quotas and limits | LiveKit docs | Understanding and Estimating Pricing for Video Conference and Livestream Us...

agents on own infra, 1000 concurrent outbound calls, 2 minutes average call duration, times 200 calls per agent i.e. 200K calls per day for 25 days a month.

Thanks — that’s enough to estimate LiveKit costs. I’ll assume agents run on your infra (so no LiveKit agent-hosting charges). Below is a concise cost breakdown and monthly estimate for the usage you provided. Given: • 1,000 concurrent outbound calls • Average call length: 2 minutes • Each agent handles 200 calls/day → 200,000 calls/day • 25 days/month Calculations 1. Participant minutes (WebRTC participant minutes): • Each call has 2 participants in the LiveKit room (caller + your agent) = 2 participants × 2 minutes = 4 participant-minutes per call. • Daily participant-minutes: 200,000 calls × 4 = 800,000 participant-minutes/day. • Monthly participant-minutes: 800,000 × 25 = 20,000,000 participant-minutes/month. • Rate: $0.0005 per participant-minute. • Cost: 20,000,000 × $0.0005 = $10,000/month. 1. Downstream bandwidth (egress): • Estimate audio bitrate for voice: choose a typical range (we’ll use 64 kbps as a mid-point). If you want a different bitrate, tell me and I’ll recompute. • Per-call data sent from LiveKit per minute = 64 kbps = 8 KB/s ≈ 480 KB/min. • Per-call egress (2 min): 480 KB/min × 2 = 960 KB ≈ 0.00094 GB. • Daily egress: 200,000 calls × 0.00094 GB ≈ 188 GB/day. • Monthly egress: 188 × 25 = 4,700 GB ≈ 4.7 TB/month. • Rate: $0.12 per GB (starting rate). • Cost: 4,700 GB × $0.12 = $564/month. 1. Optional costs (only if you enable) • Observability/recording: not included — you'd pay only if you enable recordings or observability. If enabled, recording/observability costs apply (usage-based). • Agent-hosting on LiveKit Cloud: not applicable (agents on your infra). Monthly summary estimate • Participant minutes: $10,000 • Downstream bandwidth: $564 • Optional features: depends on usage (not included) • Estimated total (LiveKit): $10,564/month Notes and next steps • I used 64 kbps for audio. If your codec/bitrate differs (e.g., 32 kbps or 96 kbps), I can recalc quickly. • This estimate assumes no recordings or additional egress (e.g., streaming). Enablement of those features will increase costs. • Exotel, STT/TTS, and LLM plugin costs are billed separately by their providers. Would you like the estimate recalculated at 32 kbps and 96 kbps, or include recording/observability cost examples?