Not free.
But cheap.
The Seat Four software costs $0.00. Running a venue on it still costs something: a small server, phone minutes and the AI that answers your leads. Here is every one of those bills, in plain numbers, and how we push each one down.
Published list prices, October 2026. Seat Four rows are built from pay-as-you-go parts; math below.
Four bills. All small. All yours.
You pay each provider directly, at its own published rate. We add nothing on top and take no cut.
One small rented VPS runs the whole platform for a single venue.
Your own Twilio number: $1.15 a month, under a cent a minute to answer calls.
Hears the caller, thinks, talks back. Phone line included in that range.
Email, text and chat replies. Free if the model runs on your own machine.
An AI phone call is four parts working together
Every AI phone agent, from any company, does the same four things. Hosted platforms bundle them into one price per minute. Seat Four lets you pick each part, which is where the savings come from.
The voice is the biggest line. That is the one worth spending on, because it is what your caller actually hears.
| Voice (text to speech) | Sounds like | Per spoken minute | Our take |
|---|---|---|---|
| ElevenLabs FlashAPI, pay as you go | Very human, quick enough for live calls | ~3¢ | Our pick |
| ElevenLabs v3 / v4API, most expressive | The most expressive ElevenLabs voices | ~6¢ | Premium |
| Cartesia SonicAPI, plan credits | Natural, made for real-time voice | ~2.8 to 3.8¢ | Strong option |
| Deepgram Aura-2API, pay as you go | Clear and professional | ~2.3¢ | Good value |
| Deepgram Aura-1API, pay as you go | Clear, a little flatter | ~1.1¢ | Budget |
| KokoroOpen source, Apache-2.0, runs on your machine | Good, fewer voices, no cloning | $0 + your GPU | For tinkerers |
Slide to your call volume
Monthly cost of an AI agent answering your phone, phone line included. Drag the slider or pick a venue size.
How the rows are built. GoHighLevel: $0.045 engine + $0.035 ElevenLabs V2.5 voice + about $0.002 AI model + $0.01165 LC Phone inbound. Unlimited: GoHighLevel AI Employee Unlimited, $97 a month per sub-account; phone charges still apply (GoHighLevel AI pricing, updated October 13, 2026). ElevenLabs Agents: $0.08 a minute + small model + Twilio inbound. Seat Four rows: Deepgram Nova-3 ears + GPT-5 mini class brain + the named voice + Twilio. Server cost not included (see hosting). Your mix will differ.
The AI brain costs pennies. The plan around it does not.
Writing a reply to a lead is cheap on every small AI model. On GoHighLevel the tokens cost about the same; what you really pay for is the subscription you need before you can use them.
| AI model (the brain) | Per 1M tokens in / out | Per reply | Per 1,000 replies |
|---|---|---|---|
| Open model on your machineQwen3 and friends on Ollama | free | $0 | $0 + power |
| GPT-5 nanoOpenAI | $0.05 / $0.40 | $0.0002 | $0.22 |
| Gemini 2.5 Flash-LiteGoogle | $0.10 / $0.40 | $0.0003 | $0.32 |
| Claude Haiku 5.5Anthropic | $0.10 / $0.50 | $0.0004 | $0.35 |
| DeepSeek FlashDeepSeek, half price off-peak | $0.30 / $1.20 | $0.0005 to $0.001 | $0.48 to $0.96 |
| GPT-5 miniOpenAI, direct | $0.25 / $2.00 | $0.0011 | $1.10 |
| GPT-5 mini inside GoHighLevelConversation AI, same token rate | $0.25 / $2.00 | $0.0011 | $1.10 + $97 plan |
Email and chat: run it local
Email replies and website chat do not need an answer in half a second. A few seconds is fine, which makes them perfect for a free open model on your own hardware.
- Ollama runs open models like Qwen3 with one command, free
- No per-message fee, ever, and your leads never leave your building
- Need more brains for a tricky one? Send just that reply to a paid model
Phone calls: pay for the voice
On a live call every half second of silence feels awkward, and callers judge you by how the voice sounds. That is where a paid voice earns its few cents.
- Keep the voice on ElevenLabs or Cartesia, so callers hear a warm, human voice
- Use a small, fast model for the brain: about a tenth of a cent a minute
- A fully local voice agent works, but needs a strong always-on GPU and tuning
What a graphics card can and cannot do
Local AI is free per message, but it runs on a graphics card (GPU), and the card's memory decides which models fit. Here is the honest picture.
Small models (8B)
Qwen3 8B is a 5.2 GB download. Writes a 300-word reply in 2 to 3 seconds on a modern card.
Mid models (14B)
Qwen3 14B is a 9.3 GB download. Noticeably better writing; a reply takes about 5 to 10 seconds.
Large models (30B class)
Qwen3 30B and 32B are 19 to 20 GB. Better at using tools like the calendar and booking links.
Your own Twilio account. Your own numbers.
Calls and texts run through a Twilio account in your name, billed by Twilio at its published rates. A SIP trunk is the cheapest way to bring calls into an AI agent: it is a plain phone line delivered over the internet.
Keep your numbers in your own Twilio account rather than buying them inside GoHighLevel. Moving later is then a settings change, not a phone number transfer.
| Twilio, US | Price | GoHighLevel LC Phone |
|---|---|---|
| Local number | $1.15 / mo | not checked |
| Inbound call, SIP trunkElastic SIP Trunking | $0.0034 / min | n/a |
| Inbound callProgrammable Voice | $0.0085 / min | $0.01165 / min |
| Outbound call | $0.014 / min | $0.0166 / min |
| Text messageper segment, plus carrier fee | $0.0083 | $0.00747 |
| Carrier fee per textAT&T, T-Mobile, Verizon | $0.0035 to $0.005 | not checked |
| Texting registration (A2P 10DLC) | $4.50 brand $15 vetting $1.50 to $10 / mo | not checked |
$5 to $15 a month for a small venue
Seat Four runs on one small Linux server. Aim for 8 GB of memory; it runs the app, the database and the automations with room to spare.
4 vCores, 8 GB, 75 GB NVMe. Solid and budget-friendly.
2 vCPU, 8 GB, 100 GB. Promo price, renews at $14.99. Easiest start.
4 vCores, 8 GB, 128 GB, before VAT. Most power for the money.
A 4 GB server works for a trial. Tight once automations grow.
No affiliate links, no referral codes, no sponsors. Prices are from each provider's site on October 14, 2026, and some depend on a 12-month term.
One venue, one month, two ways
800 minutes of AI phone calls, 1,000 AI text and email replies, 1,000 texts sent. The same work, priced on GoHighLevel and on Seat Four.
- GoHighLevel Starter plan$97.00
- Voice AI, 800 min, ElevenLabs V2.5$65.60
- Phone minutes, LC Phone$9.32
- AI replies, 1,000, GPT-5 Mini$1.10
- Texts, 1,000 segments$7.47 + fees
Number and texting registration fees not included.
- Seat Four software$0.00
- Server, 8 GB VPS$9.00
- Voice AI, 800 min, standard$23.84
- Phone minutes, Twilio$6.80
- AI replies, 1,000, GPT-5 mini$1.10
- Texts, 1,000, with carrier fees$12.50
- Number and texting registration$2.65
The honest answers
How much does an AI phone agent cost per minute?
Built yourself from pay-as-you-go parts (speech to text, an AI model, a voice and a phone line) it runs about 2 to 5 cents a minute, phone line included. Hosted platforms cost more: ElevenLabs Agents is $0.08 a minute before the phone line, and GoHighLevel Voice AI is $0.045 a minute for the engine plus $0.015 to $0.17 for the voice, plus phone charges. Prices checked October 2026.
Is GoHighLevel's unlimited AI plan worth it?
GoHighLevel's AI Employee Unlimited plan is $97 a month per sub-account, and phone charges still apply. Against a self-built voice stack at about 3.8 cents a minute it only comes out ahead above roughly 3,600 minutes of calls a month, about 60 hours. Most single venues take far fewer.
How much does an AI email or text reply cost?
Fractions of a cent. A typical lead reply costs $0.0002 to $0.003 on small models such as GPT-5 mini, Gemini Flash-Lite or Claude Haiku. Run an open model locally with Ollama and the AI part costs nothing beyond your hardware and power.
How much does hosting Seat Four cost?
About $5 to $15 a month for one venue. An 8 GB VPS is the comfortable size, for example about $8.50 a month at OVHcloud or $8.99 at Hostinger on promotion. The software itself is $0.00.
Can I run the AI on my own computer?
For email and chat, yes. Open models run on a GPU with Ollama: an 8 GB card handles 8B models, 16 GB handles 14B, 24 GB handles 30B-class models. Local models are slower and less capable than the best paid models and answer one request at a time by default. For live phone calls we recommend a paid voice, such as ElevenLabs, for now.
Free software. Fair bills. Your call.
Join the beta list to run Seat Four at your venue and see these numbers on your own invoices.
Sources, checked October 14, 2026
- GoHighLevel AI product pricing
- ElevenLabs API and Agents pricing
- Cartesia pricing
- Deepgram pricing
- OpenAI API pricing
- Gemini API pricing
- Claude API pricing
- DeepSeek API pricing
- Ollama Qwen3 model sizes
- Twilio voice pricing, US
- Twilio Elastic SIP Trunking pricing, US
- Twilio SMS pricing, US
- OVHcloud VPS
- Hostinger VPS
- netcup VPS
Prices change. Every figure is a published list price or simple math on one, for the US, and your own bill will differ. Seat Four is not affiliated with, sponsored by or paid by any company named on this page.