LiveKit Agents
livekit.com
- Free plan
- Open source
- Windows
- macOS
- Linux
- Web
- iPhone and iPad
- Android
- API
- Self-hosted
What LiveKit Agents is
LiveKit Agents is an open-source framework and cloud platform for building voice, video, and physical AI agents. Python or Node.js programs can participate in LiveKit rooms in real time, with deployment to LiveKit Cloud or a developer-chosen environment. Agents can take voice, video, and text input and return output in any modality. They support LLM tool use, frontend RPC calls, and multi-agent handoffs, and can connect to phone systems through native SIP. LiveKit Inference offers models from providers including OpenAI, Google, AssemblyAI, Deepgram, Cartesia, and ElevenLabs; open-source plugins connect to more providers. LiveKit Cloud offers session replay, transcripts, trace spans, logs, and reports. The free Build plan includes 1,000 agent session minutes, five concurrent sessions, and one deployment. Ship starts at $50/mo, with 5,000 session minutes and $0.01 per additional minute. LiveKit lists SOC 2 Type II, GDPR, HIPAA, CCPA / CPRA, and the EU–US Data Privacy Framework; PCI DSS and ISO 27001 are in progress.
Who it suits
It suits developers building realtime agents who need voice, video, text, or telephony workflows and a choice of cloud or custom deployment. Teams needing HIPAA support should note that BAAs are available for Scale and Enterprise customers.
Good to know, and what to watch
- Python and Node.js programs can join realtime rooms.
- Agents support tools, RPC calls, and handoffs.
- Native SIP connects agents to phone systems.
- Cloud provides replay, transcripts, logs, and reports.
- A free Build plan is available.
- Build includes only 1,000 agent session minutes.
- Agent Builder is intended for simple voice prototypes.
- Build support is community support.
- HIPAA BAAs are limited to Scale and Enterprise.
Plans
Every plan and priceEnterprise
Build
Ship
Scale
Free plan
Getting started
- Open the LiveKit Agents website
- Build an agent in Python or Node.js
- Connect the program to LiveKit rooms
- Choose LiveKit Cloud or a custom deployment environment
- Select the Build, Ship, Scale, or Enterprise plan
- Add model, telephony, and other integrations as needed
Limits to check
Build includes 1,000 agent session minutes, five concurrent sessions, one deployment, and $2.50 in inference credits. Ship includes 5,000 minutes and 20 concurrent sessions, then charges $0.01 per minute; Scale includes 50,000 minutes and then the same per-minute charge.
The Club's review
Overview
LiveKit Agents is an open-source framework and cloud platform for developers building realtime voice, video, and physical AI agents. It is best suited to teams comfortable building in Python or Node.js who want control over deployment and live-media workflows. Its breadth is a strength, but it is a framework to build with—not a finished assistant to configure.
Key features
Python and Node.js programs can join LiveKit rooms as realtime participants, processing voice, video, and text input and producing output in any modality. LLM tool use, frontend RPC calls, and multi-agent handoffs let teams build workflows beyond a single conversational exchange, with the corresponding need to design and implement those workflows.
Deployment can be on LiveKit Cloud or in a developer-chosen environment, giving teams a choice between managed hosting and their own setup. LiveKit Inference connects to models from OpenAI, Google, AssemblyAI, Deepgram, Cartesia, and ElevenLabs; open-source plugins extend provider options. Native SIP enables phone-system connections, including inbound and outbound calls, custom phone numbers, and human handoff.
LiveKit Cloud’s agent insights include session replay, transcripts, trace spans, logs, and session reports, useful for examining agent sessions. LiveKit says prompts, audio, and completions handled by LiveKit Inference and its model providers are not logged or retained. It lists SOC 2 Type II, GDPR, HIPAA, CCPA / CPRA, and the EU–US Data Privacy Framework; PCI DSS and ISO 27001 are in progress. BAAs are available to Scale and Enterprise customers handling protected health information.
Pricing
Build is free at $0/mo and includes 1,000 agent session minutes, five concurrent agent sessions, one agent deployment, $2.50 in inference credits, and community support. It is a useful starting point, but the session and deployment caps make it a constrained base for ongoing or multi-agent work.
Ship starts at $50/mo, billed starting at $50/mo. It includes 5,000 agent session minutes, then charges $0.01 per minute, allows 20 concurrent agent sessions and two deployments, and adds email support. This is the entry paid plan for teams that need more room than Build provides; the per-minute overage matters as usage grows.
Scale starts at $500/mo, billed starting at $500/mo. It includes 50,000 session minutes, then $0.01 per minute, up to 600 concurrent sessions, four deployments, region pinning, security reports, and HIPAA. Its larger concurrency allowance and compliance-related features target substantial workloads rather than early prototypes.
Enterprise has custom pricing and includes volume pricing, a shared Slack channel, SSO, and a support SLA. It is the option for organizations needing those terms and support arrangements. Build includes community support, Ship email support, and Enterprise a support SLA.
Platforms
LiveKit Agents supports Android, iOS, Linux, macOS, Windows, web, API, and self-hosted environments. The mix reflects its developer framework and deployment choices rather than a single end-user app.
Who it's for
Choose LiveKit Agents if your team can build and operate an agent framework and needs realtime media, telephony, provider flexibility, or a choice of cloud and custom deployment. Agent Builder is intended for simple voice-agent prototypes and produces Python code for deployment to LiveKit Cloud; it is not positioned for larger production workflows. Teams seeking a finished assistant with little development work should look elsewhere.
Pros and cons
- Pro: Realtime voice, video, and text, plus multimodal output and agent handoffs, support more than text-only conversational flows.
- Pro: Cloud or custom deployment, native SIP, and model integrations give developers several ways to shape an implementation.
- Pro: Cloud session replays, transcripts, traces, logs, and reports offer multiple views for inspecting sessions.
- Con: Building workflows in a Python or Node.js framework requires developer effort; Agent Builder is limited to simple voice prototypes targeting LiveKit Cloud.
- Con: Build caps usage at 1,000 minutes and five concurrent sessions, while Ship adds per-minute charges after 5,000 minutes.
- Con: The listed PCI DSS and ISO 27001 status is in progress, so those certifications should not be treated as completed.
Alternatives
For a broader set of voice-agent choices, browse AI Voice Agents; for platforms centered on conversational AI, see Conversational AI Platforms.
Consider Intervo if a self-hosted option with source code under an MIT License is the priority. Tone is another self-hosted choice, with its free plan specifying unlimited agents, calls, and workflows and the option to bring your own LLM, voice, and telephony.
Rapida AI may suit teams wanting a free one-agent launch plan that uses their own STT, TTS, LLM, and SIP, with 10-day observability retention. Kore.ai is a paid alternative with project-tailored custom pricing.
Ring AI offers a free tier with 100 minutes per month for testing, which may fit a small trial. Yellow.ai has a free plan specifying one AI agent, 500/month included, a two-agent limit, 1,000 records, and one active campaign.
Vapi is a paid option with usage-based billing, $5 free credits to get started, one included phone number, and four concurrent calls. Fonema is another option for call-center plans contracted through sales.
Verdict
LiveKit Agents is a strong choice for developer teams building realtime, multimodal agents that need flexible deployment, telephony, and cloud session insights. Its main trade-off is the engineering work of building on a framework, compounded by tight free-plan caps and usage-based overages on paid plans. Choose it for control over agent architecture; look elsewhere if you want a ready-made assistant or a generous free workload.
The verdict
LiveKit Agents combines realtime media, agent workflows, and deployment options in one framework and platform. Check plan limits and the scope of Agent Builder before choosing it for a larger or production workflow.
Questions people ask
Is LiveKit Agents open source?
Yes. LiveKit Agents is an open-source framework as well as a cloud platform.
What does the free Build plan include?
Build costs nothing and includes 1,000 agent session minutes, five concurrent sessions, one deployment, $2.50 in inference credits, and community support.
Can agents connect to telephone systems?
Yes. Native SIP connects agents to phone systems, with inbound and outbound calls supported.
Which programming languages are supported?
Python and Node.js programs can join LiveKit rooms as realtime participants.
Which providers are available through LiveKit Inference?
LiveKit Inference lists OpenAI, Google, AssemblyAI, Deepgram, Cartesia, and ElevenLabs, while open-source plugins connect to additional providers.
Does LiveKit offer HIPAA support?
LiveKit lists HIPAA compliance, and BAAs are available for Scale and Enterprise customers handling protected health information.
What the maker publishes
- Agent Builder
- Agent Builder is a browser-based, no-code tool included with LiveKit Cloud and is unavailable in self-hosted deployments. source
- Agent orchestration
- The documentation covers agent sessions, workflows, handoffs, tasks, task groups, tools, and external-data retrieval. source
- Client SDKs
- LiveKit offers frontend and client SDKs for web, mobile, and other platforms, while its Unity SDK supports macOS, Windows, Linux, iOS, and Android. source
- Company founded and size
- LinkedIn lists LiveKit as founded in 2021 with 51–200 employees. source
- Company headquarters
- LiveKit’s official site lists 369 Sutter Street in San Francisco, California. source
- Company legal name
- Crunchbase identifies the legal entity as LiveKit Incorporated. source
- Company location
- LiveKit lists its address as 369 Sutter St, San Francisco, California. source
- Compliance
- LiveKit lists SOC 2 Type II, GDPR, HIPAA, CCPA / CPRA, and the EU–US Data Privacy Framework; it says PCI DSS and ISO 27001 are in progress. source
- Deployment
- Agents can be deployed to LiveKit Cloud or to a custom environment of the developer’s choice. source
- Documentation
- Official documentation includes quickstarts, recipes, SDK references, API references, deployment guidance, telephony, models, testing, and troubleshooting material. source
- Encryption
- WebRTC media uses SRTP with DTLS-SRTP exchange, SIP supports TLS and SRTP, and stored data uses AES-256 encryption. source
- Headquarters
- San Francisco, California, United States source
- HIPAA
- LiveKit says BAAs are available for Scale and Enterprise customers handling protected health information. source
- Inference retention
- LiveKit says prompts, audio, and model outputs sent through LiveKit Inference are not logged or retained by LiveKit or its model providers. source
- Integrations
- LiveKit Inference provides models from providers including OpenAI, Google, AssemblyAI, Deepgram, Cartesia, and ElevenLabs, while open source plugins connect to additional providers. source
- Licence
- The Agents framework is Apache-2.0 licensed, while LiveKit turn-detection models use a separate LiveKit Model License. source
- Limit: cloud-only capabilities
- CloudTalk’s comparison says Agent Builder, LiveKit Inference, adaptive interruption handling, Krisp noise cancellation, and built-in observability are unavailable when self-hosting. source
- Limit: operational complexity
- Airframe’s review notes that self-hosting can require command-line and server-management skills and that production deployments carry operational complexity. source
- Limit: pricing concerns
- A G2 review praised capability but specifically identified pricing as a concern. source
- Media and modalities
- Agents can process realtime voice, video, and text input and produce output in any modality. source
- Model integrations
- LiveKit Inference and plugins connect agents to OpenAI, Google, AssemblyAI, Deepgram, Cartesia, ElevenLabs, and other providers. source
- Multimodal agents
- The framework supports realtime spoken interaction, video input and output, and physical-agent applications. source
- Node requirement
- The Node.js quickstart requires Node.js 20 or newer and pnpm 10.15.0 or newer. source
- Notable limit
- Agent Builder is intended for simple voice agent prototypes and produces Python code for deployment to LiveKit Cloud. source
Checked against the maker's pages on 15 Oct 2026.