What observability/telemetry data is stored in the U.S? According to https://livekit.com/legal/data-processing-addendum section 2.5 "Observability Data, telemetry, and related logs are stored and processed in the United States regardless of the selected Pinned Region;". I'd like to understand whether this would include any transcript or PII of any kind. And if so, is there a way to disable it?. Is there a way to restrict all the data to be in EU or UK?

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

If I disable collection of observability data, nothing will be sent to the U.S?

Thanks — I’m responding to your question about whether disabling Observability will stop any data being sent to the U.S. Yes. If you disable collection of Observability Data for a project or a session (for example by turning off project-level observability or setting

record=False

for a session), no observability data — including logs, traces, metrics, and Agent Observability snippets such as transcripts, audio, or video excerpts — will be collected or sent to the United States for those sessions. Important to note: • Disabling Observability only affects Observability Data. Other data types (User Content, Developer Application Data, Inference Data) remain subject to your Pinned Region settings and any region restrictions you’ve configured. • Observability Data is otherwise always processed/stored in the U.S. if collection is enabled. If you’d like, I can: • Walk you through disabling Observability at the project level, or • Show the exact place to set

record=False

in your agent/session code. Which would you prefer?

so that would disable the "Agent insights" tab in Livekit cloud, correct?

Thanks — I’m responding to your question about whether disabling Observability will disable the "Agent insights" tab in LiveKit Cloud. Yes. Disabling observability data collection for a project or a session (for example, turning off project-level observability in Data & Privacy or setting

record=False

for an agent/session) prevents collection of the transcripts, logs, traces, and audio/video snippets that Agent Insights uses. As a result, the "Agent insights" tab will not show session data or the session-replay / turn-by-turn analysis it normally provides. A couple of quick reminders: • Disabling observability only affects Observability Data. Other data types (User Content, Developer Application Data, Inference Data) remain governed by your Pinned Region settings. • When observability is disabled, no observability data is collected or sent to the U.S. for those sessions. Would you like step‑by‑step help to disable observability at the project level or to set

record=False

in your agent/session code? Sources: LiveKit Agent Observability | Debug and optimize voice agents | LiveKit | Streamline troubleshooting with Agent Observability | LiveKit

If I deploy the livekit agent to livekit cloud, will it be in the pinned region?

Thanks — I’m responding to your question about whether an agent you deploy to LiveKit Cloud will run in your pinned region. Yes. When you deploy a LiveKit agent to LiveKit Cloud and you have configured region pinning, the agent and the developer application data run in the pinned region you selected. User content (real‑time audio, video, text, and session metadata) is routed and processed within that same pinned region. Important caveat: • Observability data (telemetry, logs, traces, and Agent Observability snippets such as transcripts, short audio/video excerpts, and synchronized session playback) is always stored and processed in the United States regardless of your pinned region. That means even though the agent runs in your pinned region, any enabled observability collection will still send that observability data to the U.S. If you want to avoid observability data going to the U.S., you can disable observability: • At the project level in LiveKit Cloud (Data & Privacy), or • Per session/agent by setting

record=False

in the agent/session code. Would you like step‑by‑step guidance to confirm your current pinned region, or help disabling observability at the project or session level? Sources: Data Processing Addendum | LiveKit | Streamline troubleshooting with Agent Observability | LiveKit | LiveKit Agent Observability | Debug and optimize voice agents | LiveKit

Can I use livekit inference and guarantee the data stays in the pinned region?

Thanks — I’m responding to your question about whether LiveKit Inference can be guaranteed to keep all data inside your pinned region (for example, EU or UK). Short answer: No — you cannot guarantee that Inference data will stay in your pinned region unless the inference provider you use actually runs infrastructure in that same region. Why: • Region pinning ensures LiveKit’s media, signaling, and developer application data run in the pinned region. • Inference Data (model inputs like prompts or audio snippets, metadata, and model outputs) is sent to whatever inference provider you select. If that provider’s infrastructure is outside your pinned region, the data will be processed outside your pinned region. • Many common inference sub‑processors (for example, several listed by LiveKit) operate in the United States, so using them will send inference data outside the EU/UK. What you can do: • Choose an inference provider that has infrastructure in your desired region (EU/UK). If that provider processes data inside the region, your inference traffic can stay there. • Audit the provider’s regional availability and data handling policies before connecting them. • If you need, I can help identify which providers LiveKit currently supports and which of those have regional deployments (EU/UK) so you can pick one that meets your residency requirements. Would you like me to look up the list of supported inference providers and note which ones have EU/UK infrastructure, or walk through how to configure a region‑local provider with LiveKit? Sources: Data Processing Addendum | LiveKit | Sub-processors | LiveKit

Yes please

Great — I’ll help you find inference providers that can keep data in the EU/UK and show options to achieve EU/UK residency. Quick recap of the requirement you confirmed: you want to use LiveKit inference while keeping inference data (prompts, audio snippets, metadata, and model outputs) inside your pinned region (EU or UK). A few important facts up front: • LiveKit’s region pinning keeps media, signaling, and agent code inside the pinned region, but inference data is processed by whichever inference provider you choose. • If the inference provider’s infrastructure is outside the EU/UK, your inference data will leave the pinned region. • Many commonly used inference sub‑processors listed by LiveKit operate in the United States, so using them will not keep data in the EU/UK. I can help in two concrete ways — which would you like me to do first? 1. Look up the list of LiveKit-supported inference providers and identify which (if any) have EU or UK infrastructure that would keep inference data in-region. 2. Walk you through options to keep inference data in EU/UK even if LiveKit’s default providers don’t: ◦ Self-hosting STT/LLM/TTS in your EU/UK infrastructure and calling them from your agent, or ◦ Connecting to a third‑party provider that explicitly guarantees EU/UK processing (and reviewing their regional endpoints/configuration). If you choose option 1, please confirm whether you need EU (continental) or UK only, and whether you need specific model types (STT, LLM, TTS, or all). That will help me filter providers that meet your needs. If you choose option 2, I can provide step‑by‑step guidance on: • Deploying an agent to the EU region (example

eu-central

), and • Configuring the agent to call a self‑hosted inference endpoint so inference data never leaves your infrastructure. Which path would you like to take? Sources: Data Processing Addendum | LiveKit | Streamline troubleshooting with Agent Observability | LiveKit | Streamline troubleshooting with Agent Observability | LiveKit | Agents on LiveKit Cloud | Deploy voice AI on global infrastructure | LiveKi... | Sub-processors | LiveKit

option 1, EU or UK any is fine

Great — I’ll list LiveKit-supported inference providers that can keep inference data in the EU/UK and explain what you need to do to keep inference data in‑region. Quick summary • You can keep inference data in the EU/UK only if the inference provider you use runs infrastructure in that region. LiveKit will route inference calls to the chosen provider’s endpoints. • Providers confirmed (or strongly indicated) to have EU/UK infrastructure: Google and Deepgram. Using these with EU/UK region endpoints can keep inference traffic in‑region. • Many other commonly used providers (AssemblyAI, some others listed as sub‑processors) operate in the U.S., so using them will send inference data outside EU/UK. • Observability (logs, traces, transcripts used by Agent Insights) is always stored in the U.S. unless you disable observability (project-level or per-session with

record=False

). Providers you can consider for EU/UK residency • Google ◦ Has EU and UK data centers. If you select Google models and configure the EU/UK endpoint for inference, your inference data can remain in‑region. • Deepgram ◦ Offers EU-based inference endpoints and models that can be routed to EU infrastructure. • (Notes) ElevenLabs and other providers headquartered in Europe/UK may have regional hosting, but you must confirm region-specific endpoints and commitments with each provider before use. What you must do to keep inference data inside EU/UK 1. Choose a provider with explicit EU/UK endpoints (for example Google or Deepgram). 2. Configure your agent or inference settings to use the provider’s EU/UK endpoint (region-specific model IDs or endpoint URLs). 3. Verify with the provider that the specific model and endpoint you use are hosted in the EU/UK and that they don’t route or replicate data to the U.S. 4. Disable LiveKit Agent Observability if you require that no transcripts/observability snippets be sent to the U.S. (project settings → Data & privacy, or set

record=False

per session). Practical recommendations • If you need a short prescriptive path: pin your project/agent to an EU region (e.g., eu-central), deploy the agent there, choose Deepgram or Google with EU endpoints for STT/LLM/TTS as needed, and set

record=False

to prevent observability traces/transcripts from being collected. • If any doubt remains about a provider’s regional guarantees, ask the provider for a written statement about regional data residency or for an enterprise contract clause that restricts processing to EU/UK locations. Would you like me to: • Pull the exact model IDs / endpoint names for Google or Deepgram that LiveKit routes to in the EU, or • Walk through the steps to configure an agent to use Deepgram/Google EU endpoints and set

record=False

? Sources: Data Processing Addendum | LiveKit | LiveKit Agent Observability | Debug and optimize voice agents | LiveKit | Streamline troubleshooting with Agent Observability | LiveKit | Sub-processors | LiveKit | Deepgram STT | LiveKit Documentation

can i specify the endpoints if I use livekit inference?

Thanks — I’m responding to your question about whether you can specify custom endpoints when using LiveKit Inference. Short answer: No — you cannot provide custom endpoints when using LiveKit Inference. LiveKit Inference routes requests internally based on the model ID you configure; you cannot override that routing with a custom base URL or endpoint. What this means practically: • Pick a model ID that corresponds to a provider and region (for example, Deepgram or Google models). LiveKit will route to the provider’s endpoint for you. • You cannot point LiveKit Inference at your own custom or self‑hosted inference endpoint. If you need to use a custom endpoint (for example, an OpenAI‑compatible or self‑hosted API), use a plugin or call the provider directly from your agent code instead of using the

inference.*

wrappers. For example: • Use an LLM/STT/TTS plugin or SDK that lets you set

base_url

and API key to your custom endpoint, or • Make direct HTTP requests from your agent to your self‑hosted inference services. If you’d like, I can: • Show how to switch from

inference.LLM

to a plugin or direct call (with a short code example), or • Help pick model IDs that LiveKit will route to EU/UK endpoints (e.g., Deepgram/Google) so inference stays in‑region where supported. Which would you prefer? Sources: Introducing LiveKit Inference: A unified model interface for voice AI | Liv... | Large language models (LLM) overview | LiveKit Documentation | Deepgram STT | LiveKit Documentation | Models overview | LiveKit Documentation