LiveKit Dashboard
Community Get in touch
Self-hosted · LiveKit admin panel · v1.38

Stop Paying Per Minute.
Keep the Profits.

The self-hosted LiveKit admin panel that replaces command lines with clicks. Build, launch and manage AI voice agents that answer and make real phone calls — running on infrastructure you own, storing data on your server, with client logins under your brand. No coding required.

No per-minute platform fees Runs on your server Geo-located AI models White-label client portals No code needed
Get in touch Join the community See what it does
dashboard.yourdomain.com

Overview

Live performance across every agent and call.

User Guide? HelpAdministrator
OverviewCalls & RoomsLatencyAI InsightsAgentsTelephonyPrice/Cost
ROOM CONNECTION RATE98.2%
ANSWER RATE94.6%
CALLS TODAY128
RUNNING AGENTS6
TOTAL CALLS1,284
TOTAL MINUTES3,412
AVG DURATION2m 39s
AVG SENTIMENT0.82
Call Volume · last 12 hours
Sentiment
  • Positive 82%
  • Neutral 14%
  • Negative 4%
Click any page on the left, or the panel itself, to exploreTronic LiveKit Dashboard © 2026
$0
Per-minute platform fees
100%
Your data, your server
GUI
No code needed
White-label
Client portals built in
The problem

Why I built this

If you are building voice AI in the US, life is easy. For the rest of us — Australia, Europe, Asia — we are fighting physics.

Before your AI even thinks, you have lost 300ms sending audio to an American server. Add the processing time on top and the conversational magic breaks. The caller talks over the agent, the agent talks over the caller, and the demo that felt like magic on a US laptop feels broken on a real phone line.

  • 300ms+ of latency before AI processing even begins
  • Managing LiveKit servers over SSH and command lines
  • Digging through log files to debug voice agent issues
  • No visual interface for agent management
  • Per-minute platform fees eating the margin on every client

I needed to host locally but manage globally. So I built the tool I needed — then made it something a non-technical operator could run.

Round-trip latency

Audio transit time, before AI processing
US caller → US server~50ms
AU caller → US server~350ms
AU caller → AU server~50ms

Host the server near your callers and select on-shore models — AWS Bedrock in Sydney, Google Vertex AI in Australia, Azure in your region — and the transit penalty disappears before the model has done any work.

Most voice-AI products rent you an agent on their cloud. This one gives you the whole factory — installed on your server, answering on your numbers, storing your data.

What sets it apart

Why this dashboard is different

A stateless, ultra-lightweight control panel that gives you full GUI control without the database bloat — and without adding a millisecond to the call loop.

Engineered for non-US latency

Select geo-located models — Gemini in Sydney, Bedrock in London — to keep processing on-shore and shave 250–300ms of transit off every turn. The stateless architecture adds zero latency to the call loop.

Goodbye command line

A visual editor for prompts, temperature, voices and models. Hot-swap configuration while agents are running. Live debugging shows real-time transcription and tool usage as the call happens.

Smart Flux integration

Deepgram Flux solves the interruption problem. Turn-complete events understand context, so your agent waits politely when a caller pauses mid-sentence instead of talking over them.

Built for agencies

White-label sub-users for clients. They see your logo, their agents and their recordings — nothing else. You keep admin control, plus cost tracking for real-time margin analysis.

You can see everything

Live call monitoring, full history with recordings and transcripts, AI-written summaries, analytics, and an itemised cost breakdown for every single call — so you always know what your agents did and what it cost.

It tests and improves itself

Talk to an agent before it goes live with the built-in tester. Run scripted AI-vs-AI simulations at scale. Then let the Improvement Lab replay real calls against different settings to find what actually works better.

Latency X-Ray

See exactly where every millisecond goes

Speed is the metric that decides whether a call feels like a conversation or a phone tree. So the dashboard measures it properly — P50, P90 and P99 response times, broken down by speech-to-text, language model and text-to-speech, with a per-agent table underneath.

When an agent feels sluggish you do not guess. You open the breakdown, see which stage is eating the budget, and change that one provider.

  • Component breakdown — STT, LLM and TTS timings separated, not averaged into one useless number
  • Latency trend — 24-hour history, so you can see whether a provider degraded overnight
  • Per-agent detail — average TTFB, P50 and P90 for every agent, with sample counts
  • Industry targets built in — under 500ms natural, 500–1000ms acceptable, over 1000ms a noticeable pause
dashboard.yourdomain.com/analytics/latency
Latency screen showing API latency, average time to first byte, P50, P90 and P99 metrics, a component breakdown donut for STT, LLM and TTS, a 24-hour latency trend chart and a per-agent latency table
Response latency metrics with the STT / LLM / TTS split and a per-agent table.
Any provider, any region

Pick the brain, ears and voice of every agent

Works with OpenAI, Anthropic, Google, AWS, Azure, xAI, Deepgram, ElevenLabs, Cartesia and more. Mix and match per agent, and swap providers in two clicks when a better or cheaper model appears.

The part that matters outside the US: models are tagged by region. Choosing AWS Bedrock (Geo-Located) with Claude in the APAC region, or Vertex AI in Australia, keeps the processing on-shore instead of round-tripping to Virginia.

  • Geo-located model tagging — on-shore endpoints marked separately from global ones
  • Two pipeline modes — classic STT → LLM → TTS, or ultra-low-latency realtime audio models
  • Per-agent configuration — a cheap model for the switchboard, a strong one for the specialist
  • No vendor lock-in — you hold the API keys, and you can change your mind at any time
dashboard.yourdomain.com/agents/edit
Agent behaviour tab showing the language model provider set to AWS Bedrock Geo-Located with Claude Sonnet 4 in the APAC region, an LLM temperature slider and conversation dynamics settings
Model selection with region-tagged providers — here, Claude on Bedrock in APAC.
Agent control centre

Build an agent in an afternoon, not a sprint

Stop managing agents over SSH. Start, stop, edit and duplicate them with a click. Write the agent’s instructions in plain English, give it a greeting, load your business facts into its knowledge base and tell it which tools it may use.

Change a system prompt while the agent is running and it picks the change up on the next call — no restart, no deployment, no downtime.

  • No-code management — a visual interface for every agent operation
  • Hot-swapping — change prompts and tool definitions while the agent is live
  • Smart handoffs — route a caller to another agent with full context, a summary, or a fresh start
  • Deepgram Flux — native turn-complete events for phone numbers, emails and other slow dictation
  • Agent memory — agents remember returning callers and previous conversations
dashboard.yourdomain.com/agents/edit
Agent editor identity tab with fields for agent name, slug, initial greeting, system prompt, knowledge base and tool usage instructions
The agent editor — greeting, system prompt, knowledge base and tool instructions in plain English.
Agency & client management

Give clients a login. Keep the admin panel.

Built for agencies running voice AI for other businesses. Create a sub-user for each client and they get their own login showing their agents, their recordings and their analytics — under your logo, not the system backend.

Enter your own LLM, STT and TTS token costs and the dashboard calculates margin in real time, so you know what a client actually costs you before you send the invoice.

  • White-label sub-users — clients see your brand and only their own data
  • Granular permissions — restrict a client to view-only, or to specific agents and phone numbers
  • Cost tracking — per-call, per-agent and per-day cost with margin analysis and billing reports
  • Multi-tenant ready — run every client on one install, or give each their own
dashboard.yourdomain.com/analytics/calls
Calls and rooms analytics showing total calls, answered calls, answer rate, total minutes, average duration, a call volume trend chart and call outcome breakdowns
The reporting a client actually wants to see — answer rate, volume trend and outcomes.
White-label

Your name, your logo, your emails

There is a White-Label tab in settings. Upload your logo and favicon, set the dashboard name, and the product stops looking like someone else’s software. Your clients see your brand in the sidebar, in the browser tab, and on every email the system sends.

Nothing in the interface says Tronic or LiveKit unless you want it to — and it is a genuine setting, not a paid unlock or a custom build.

  • Custom logo — around 200x50, PNG, JPG or SVG, swapped in a single upload
  • Custom favicon — the browser tab icon, 32x32 or 64x64, PNG, JPG, SVG or ICO
  • Dashboard name — sets the browser tab title, so it reads “Settings — Your Dashboard”
  • Branded email — login links, reports and alerts all go out under your name and address
  • Your own domain — run the whole thing on dashboard.yourdomain.com
  • One-click reset — drop the custom branding and fall back to the default whenever you like
dashboard.yourdomain.com/settings/white-label
White-Label Branding settings showing a custom logo upload with size guidelines, the current branding preview, a custom favicon upload, and a dashboard name field with a live browser tab preview
Settings → White-Label — logo, favicon and dashboard name, with a live preview of the browser tab.
Real-time observability

Watch a live call, not a log file

Click into any active room and see it happening: who is connected, which tracks are live, what the caller just said and which tool the agent just called. When something goes wrong you watch it go wrong, instead of reconstructing it afterwards from a log.

Finished calls keep their room visible for several minutes afterwards, so a call that ended badly is still there when someone comes to tell you about it.

  • Live room view — participants, tracks, RTC connection stats and join times
  • Ghost rooms — rooms persist after the call ends so events can still be debugged
  • Server health monitor — resource tracking for the dashboard, LiveKit, Redis and SIP services
  • AI insights — auto-generated summaries and outcome classification for every call
  • Full call history — recordings, word-by-word transcripts, health status and itemised cost
dashboard.yourdomain.com/rooms
Live room detail view showing participant count, maximum participants, created time, empty timeout, RTC connection stats and a participants table with the SIP caller and the agent
A live room mid-call — the SIP caller and the agent, with RTC stats on demand.
Infrastructure & security

Hardened by default, managed from the GUI

A server on the public internet gets scanned within minutes of going live. The dashboard ships with Fail2ban wired into the interface, so the IPs probing your box for WordPress exploits get banned automatically — and you can see exactly who was banned and why.

Whitelist your office IP with one click so you can never lock yourself out, and keep the software current from the same screen.

  • Integrated Fail2ban — automatic bans for IPs scanning for vulnerabilities, managed from the GUI
  • IP whitelisting — add trusted sources and SIP trunk IPs in one click
  • Hybrid cloud storage — keep the dashboard local, offload recordings to your own S3 or GCS bucket
  • One-click updates — see which packages have updates and install them from the dashboard
  • Backup and restore — built in, not a cron job you have to remember to write
  • Audit logging and role-based sub-users — staff see only what they need to
dashboard.yourdomain.com/settings/security
IP protection settings showing Fail2ban protection active, counts of banned IPs, whitelisted IPs and active jails, tables of banned and whitelisted addresses, and a recent security events log
IP protection — live bans, whitelist and the scanner events that triggered them.
Everything included

The complete feature set

Everything below is included in the paid licence — no add-on for analytics, no seat count, no tier that withholds telephony. The free community edition runs the same software with a reduced feature set and a cap of two voice and two chatbot agents.

Build voice agents

Create · Configure · Connect knowledge

  • Visual agent editor — instructions, personality, greeting, voice and language, with no code
  • Two pipeline modes — classic speech-to-text to AI to text-to-speech, or ultra-low-latency realtime audio models
  • Knowledge base — upload your documents so agents answer from your own business information
  • Agent memory — agents remember returning callers and past conversations
  • Multi-agent handoffs — a receptionist can silently hand a caller to a sales or support specialist mid-call
  • Turn-taking controls — noise cancellation, background office ambience, interruption sensitivity and VAD timing

Put them on the phone

Telephony · Numbers · Campaigns

  • Guided SIP trunk wizard — connect your phone provider without networking expertise
  • Inbound routing — point any number at any agent, with business-hours time profiles
  • Outbound campaign dialler — upload a list, pick an agent, and it works through the list for you
  • Warm transfer to a human — the agent briefs your staff member privately before connecting the caller
  • Call recording — stored locally or pushed to your own S3 or GCS bucket

Tools & integrations

Function calling · Webhooks · Automation

  • Built-in tools — hangup, hold, play audio, get date and time, attended and blind transfer
  • Bookings — agents can check availability and book appointments through Cal.com
  • Webhook tools — call any HTTP endpoint you own, with a tester and per-tool logs
  • Automation platforms — trigger n8n and GoHighLevel workflows straight from a call
  • Messaging — send an SMS or an email mid-call or afterwards
  • API access via MCP — let your own tools and AI assistants manage the dashboard programmatically

Chat on your website too

Web widget · Text & voice

  • Embeddable chat widget — the same AI brains answering on your website, in text or voice
  • Fully brandable — colours, position and welcome messages under your own identity
  • Shared knowledge and tools — the widget and your phone agents give the same answers
  • Chat history — every session logged with duration, outcome, cost and full transcript

Monitor, measure, improve

Visibility · Testing · Reporting

  • Live view of active calls and rooms as they happen
  • Call history with audio playback, word-by-word transcripts and AI summaries
  • Per-call cost breakdown by provider — your exact cost per call, per agent, per day
  • Automated reports — daily and weekly summaries, or per-call reports, to your inbox or a webhook
  • Agent Tester and simulation — talk to an agent yourself, or run scripted AI-vs-AI tests at scale
  • Improvement Lab and AutoResearch — replay real calls against different settings and get prompt suggestions

Run it with confidence

Operations · Security · Scale

  • One-click self-updates — fixes and new features arrive through the dashboard itself
  • Backup and restore built in
  • Security hardening out of the box — login protection, IP controls, audit logging, Fail2ban
  • Multi-server clustering — add capacity nodes as call volume grows
  • Role-based sub-users — staff and clients see only what they need
  • Health checks that tell you exactly what is misconfigured, in plain English
dashboard.yourdomain.com/tools
Tools screen showing available tools, webhook tools, categories and agents with tools, an agent tool assignment table, and a tool registry with send email, send SMS, attended transfer, blind transfer, hangup and hold
The tool registry, and which tools each agent is allowed to use.
dashboard.yourdomain.com/agents/edit
Agent voice and speech tab with speech-to-text provider and model, interrupt sensitivity slider, and text-to-speech provider, model and voice selection
Voice and speech configuration — STT, interrupt sensitivity and TTS voice.
Smart Flux integration

The interruption problem, solved

Older agents rely on a silence timer: stop talking for 700ms and the agent assumes you are finished. Real people pause in the middle of an email address. Deepgram Flux understands context instead of counting silence.

Silence-timer agent
Caller: “My email is peter…”
[ pause ]
Caller: “…dot smith…”
Interrupts.“Got it! So your email is peter@…” — and now the caller has to start again.
Flux-enabled agent
Caller: “My email is peter…”
[ pause ]
Caller: “…dot smith…”
Waits politely.Turn-complete events tell it the sentence is not finished, so it lets the caller land the whole address.
Self-hosted vs managed cloud

What changes when you own the stack

Managed voice-AI clouds are excellent products. They are also a rented seat, priced per minute, in someone else’s country.

 Managed voice-AI cloudLiveKit Dashboard
Platform pricingPer minute, forever, on top of model costsA flat licence. You pay providers directly at cost
Where calls are processedTheir region, usually the USYour server, in the region you choose
Where recordings liveTheir storageYour disk, or your own S3 / GCS bucket
Model choiceTheir supported listAny provider you hold a key for, region-tagged
Client accessTheir branding, or no client access at allWhite-label sub-users under your logo
Margin visibilityTheir invoice, after the factPer-call cost and margin, in real time
If they change the pricingYou absorb it or migrateNothing changes. It is your server
Setup effortSign up and goOne install, with setup support included

Honest version: a managed cloud is faster to start and someone else carries the uptime. Self-hosting wins on cost at volume, on latency outside the US, and on anything where the data cannot leave your control.

Who it is for

Four kinds of people run this

Agencies & resellers
Deliver voice AI to clients on infrastructure you control, with each client on their own login and their own data — and a real margin instead of a per-minute markup.
Business owners
Never miss a call again. An AI receptionist that answers 24/7, books appointments and costs a fraction of a staffed phone line.
Privacy-conscious orgs
Healthcare, legal and finance — anyone who cannot ship call recordings to a third-party SaaS. Self-hosting keeps everything in-house and in-country.
Teams already on LiveKit
A ready-made management layer for the open-source LiveKit stack, instead of building dashboards and telephony plumbing from scratch.
Pricing

A licence, not a meter

There is a free version you can run forever, capped at two voice agents and two chatbot agents. The paid tiers lift the caps and unlock the whole dashboard — all agents, all telephony, all analytics, unlimited calls. What changes above that is how much of the running you want to hand over.

Free

Community edition
$0forever
  • Up to 2 voice agents
  • Up to 2 chatbot agents
  • Self-hosted on your own server
  • Full community access, setup help included
  • Limited feature set — see the note below
  • No card required
Join the community
Most popular

Premium

Self-hosted licence
$49/ month
  • Unlimited voice and chatbot agents
  • Full white-labelling enabled — your logo, favicon, dashboard name and emails
  • Every feature unlocked
  • Installation support
  • Classroom access — the full course library
  • Direct WhatsApp access for help
  • No seat count, no call limit
Get started

VIP

Done for you
$199/ month
  • Everything in Premium
  • Fully hosted for you on AWS Lightsail, in your region
  • Maintenance and updates handled
  • Ongoing management and monitoring
  • You use the dashboard; we run the box
Talk to us

The free community edition runs on your own server with a reduced feature set and a limit of two voice agents and two chatbot agents — enough to run a real phone line and prove it works. Premium removes the caps and unlocks everything. On Premium you supply the server; on VIP the hosting is included. AI provider usage is billed by the providers, directly to you, at their cost — there is no per-minute platform fee on any tier. Prices are per month and can be cancelled at any time.

FAQ

Questions people actually ask

What exactly am I buying?

A licence for the dashboard software, installed on a server you control. There is also a free community edition, capped at two voice agents and two chatbot agents, with a reduced feature set. It is the complete admin panel for LiveKit: agents, SIP telephony, recordings, transcripts, analytics, testing and security — all through a web interface. You are not renting agent minutes from us; you are running the platform yourself.

What is in the free version?

The community edition is the same software, self-hosted on your own server, with a reduced feature set and a cap of two voice agents and two chatbot agents. That is enough to put a real AI receptionist on a real phone number and see how it performs before paying anything. Upgrading to Premium lifts the caps and unlocks the rest of the dashboard — you keep the same install and the same data.

What do I need to run it?

A Linux server — a modest cloud VM is enough to start — a domain name, and API keys for whichever AI providers you want to use. Installation support is included with the licence, so you do not have to work it out alone. If you would rather not run a server at all, the VIP tier hosts it for you.

Do I have to be technical?

Not after it is installed. Everything — agents, phone lines, monitoring, updates, security, backups — is point and click. There is an in-app user guide with screenshots, health checks that name what is misconfigured in plain English, and one-click updates. No terminal, no code and no IT department for day-to-day operation.

Is there a per-minute fee?

Not from us. You pay a flat monthly licence, and you pay your AI and telephony providers directly at their own rates. Nobody takes a cut of every minute in between. That is the entire economic argument for self-hosting: at volume, the per-minute platform fee is usually larger than the model cost it sits on top of.

Can I put my own brand on it for clients?

Yes. White-labelling is built in. Create a sub-user for each client and they log in to your domain, under your logo, seeing only their own agents, recordings and analytics. Permissions are granular, so a client can be view-only or restricted to specific agents and phone numbers.

Which AI providers can I use?

OpenAI, Anthropic, Google, AWS Bedrock, Azure, xAI, Deepgram, ElevenLabs, Cartesia and more, mixed and matched per agent. Regional endpoints are tagged separately, so you can deliberately keep processing on-shore — Bedrock in Sydney, Vertex AI in Australia, Azure in your region — instead of round-tripping to the US.

Where does my call data live?

On your server. Recordings, transcripts and customer details are stored where you put them — locally, or in your own S3 or Google Cloud Storage bucket. Nothing is warehoused with us. For healthcare, legal and finance work, that is usually the difference between a project that can proceed and one that cannot.

How is this different from Tronic Cloud Voice AI?

They are two different products from the same team. Tronic Cloud Voice AI is a hosted, pay-as-you-go platform — you sign up, get an Australian number and build agents on our infrastructure. LiveKit Dashboard is the self-hosted option: you own and run the whole platform, brand it as your own, and create accounts for your own clients. Hosted is faster to start; self-hosted is what you want when the infrastructure, the data or the margin has to be yours.

How do I get help?

Through the community, which is free to join and where setup questions get answered, and directly over WhatsApp on the Premium and VIP tiers. Installation support is included with the licence. You can also email voiceai@tronic.com.au at any time.

Join the high-performance voice AI community

Do not settle for demos that only work on a US laptop. Build infrastructure that holds up on real networks, for real clients, at real volume — and keep the margin while you do it.

Get in touch Join the community