For your team
Genie Chat
Drafts, summaries, analysis and code for everyone in the office. It reads your own files, and nothing it sees is used to train anyone else's model.
From $10a seat a month, or included on your box3
One box runs Genie for your team, your website and your phone line. It stays in your building, works without the internet and never sends you a token bill. Plug it in: we tune the hardware and software together, and we look after it. Three sizes: Box Lite for chat and documents, Box Pro adds voice and images, Box Enterprise adds video and more users.

Get one for your website:
How it works
The Superchat Box is your company's AI, running on your desk. Your team asks it questions, your customers get answers on your website, and your callers talk to it on the phone. Same box. Same private data. No meter running.
For your team
Drafts, summaries, analysis and code for everyone in the office. It reads your own files, and nothing it sees is used to train anyone else's model.
From $10a seat a month, or included on your box3
For your customers
An assistant on your website that answers from your own products, prices and policies, day and night, and turns interested visitors into leads.
From $20a month. One line of code on your website
For your callers
A voice agent for your phone line or front desk. It listens, understands and answers out loud, in English and Hindi.
From $50a month. Real-time voice in English and Hindi
Chat and Live Support run on Box Lite. All three run on Box Pro
Try it now
No sign-up. Pick one and use it.
Superchat Box
Cloud AI bills for every word it reads and writes. Move the slider to see a year of that bill. On your own box it is zero: Box Lite is from $5,500, Box Pro from $20,000 and Box Enterprise from $50,000, once.
Estimate for 0 tokens a year at a blended $10 per million tokens and 22 working days a month. A Superchat Box is a one-time price: from $5,500 (Lite), from $20,000 (Pro) or from $50,000 (Enterprise).
Genie Chat
These are live answers from Genie. Ask about your work, have it draft something, or test it with a hard question.
Genie Live Support
This is the assistant on our own website. Ask about products or prices, or ask for a person, and see how it handles you.
Powered by Genie
Genie Orb
Tap the mic, allow your microphone and speak in English or Hindi. It answers out loud.
Inside every product
Lite answers in an instant and runs on every box. Pro takes on the questions that need real thought, on Box Pro.
Fast
Built for speed: everyday questions, website conversations and spoken replies that start the moment you stop talking.
120B-equivalent
Our most capable model, for reasoning, long documents, analysis and code. No data-centre GPU needed: it runs on the box itself.
48tokens per second on a Superchat Box1
Why Superchat
Cloud AI is easy but your data leaves and the bill never stops. Building your own is private, but it means data-centre GPUs, engineers and a network to run them. Superchat gives you both halves.
| Question | Cloud AI services | Build it yourself | Superchat |
|---|---|---|---|
| Where your data goes | To the provider's servers | Stays with you | Stays in your building |
| How you pay | Per seat or per token, every month | Hardware, plus your engineers' time | Once for the box. No token bill. |
| What you have to buy | Nothing, but you rent it for good | Data-centre GPUs, servers and networking | One box. Plug it in. |
| Hardware and software | Not yours | Pieced together from different vendors | Tuned together by us |
| When the internet is down | Stops working | Keeps working | Keeps working |
| Time to the first answer | Minutes | Weeks of setup | About 15 minutes |
| Website support and voice | Separate products to buy | Yours to build | Included |
| Updates and maintenance | From the provider | Your engineers maintain it | From Superchat, with an annual maintenance contract |
Start today
Most customers start with one product and add the rest. Everything here is available now.
Genie Live Support
Sign up, add your site and paste one line of code.
From $20a month. Sign up online and start today
Genie Chat
Hosted for you, or running on your own box.
From $10a seat a month3
Superchat Box
Box Lite for private chat. Box Pro adds Genie Pro, voice and image. Installed and ready.
From $5,500One time, no token bill. Box Pro from $20,000, Box Enterprise from $50,000. Ships with Ubuntu, Genie OS from March 2027.
Genie Orb
We set it up with you, on your calls and in your languages.
From $50a month. We set it up with you
In production
Eight organisations across banking, healthcare, aerospace, retail, genomics, IT and education. Names under NDA, references on request.
We bring a box to your office, load your documents and pull the network cable while it answers.
Tell us where it's going. Our sales team will get back to you.
Your email is confirmed. Our sales team will get back to you shortly about .
Genie's models next to two well-known flagship models. Scores in percent, as of 10 November 2026.
| Benchmark | GPT-5OpenAI | Claude Opus 4.5Anthropic | Genie CortexSuperchat Genie | Genie Lite (free)Superchat Genie |
|---|---|---|---|---|
| GPQA Diamond | 85.7 | 87.0 | 84.2 | 32.6 1 |
| HLE, no tools | 24.8 | 30.8 | 22.4 | – |
| HLE, with tools | 35.2 | 43.2 | 47.4 | – |
| HMMT Feb 2025 | 93.3 | 92.9ᵗ | 89.0 | – |
| AIME 2025 | 94.6 | 93.0 | 93.3ᵗ (AIME 2026) | – |
| SWE-bench Verified | 74.9 | 80.9 | 69.2 | – |
| SWE-bench Pro | 41.8ᵗ | 45.9ᵗ | 68.6ᵗ (v2) 2 | – |
| Terminal-Bench 2.0 | 49.6ᵗ | 59.3 | 40.5 | – |
| OSWorld | – | 66.3 | 54.5 (Verified) | n/a |
| BrowseComp | 54.9 | 67.8 | 61.0 | – |
| MMMU (vision) | 84.2 | 80.7 | 81.4 | n/a |
| MMMLU (multilingual) | – | 90.8 | 85.2 | 61.9 |
| MMLU-Pro | 86.5ᵗ | 89.5ᵗ | 85.3 | 44.9 |
| IFEval | 92ᵗ 3 | 90.9ᵗ | 91.9 | 81.4 |
| LiveCodeBench v6 | – | – | 74.6 | – |
Plain number = declared by the model's maker. ᵗ = measured by a third party (sources below), used where the maker published nothing. n/a = text-only model (vision benchmark). – = no published figure we could find.
The GPT-5 and Claude Opus 4.5 scores are published numbers we gathered from the internet: the makers' own announcements and public leaderboards. We haven't run those models ourselves.
The Genie scores are the published scores of the base models Genie is built on. We test Genie internally. Genie runs compressed 4-bit builds, mostly without extended thinking, so its real scores can be lower: our own GPQA Diamond run scored 41.9.
Test settings and versions differ from source to source, so treat this as a guide, not a like-for-like comparison. Everyone is free to run their own tests.
Versions: Terminal-Bench 2.0 and OSWorld-Verified are shown as such; other versions of these tests aren't comparable.
Third-party sources (ᵗ): vals.ai (GPT-5 MMLU-Pro); Artificial Analysis (Claude Opus 4.5 MMLU-Pro); Scale SEAL, SWE-bench Pro public set (GPT-5, Claude Opus 4.5); tbench.ai leaderboard (GPT-5 Terminal-Bench 2.0, with Codex CLI); MathArena (Genie Cortex AIME 2026); a model maker's comparison table via benchlm (Claude Opus 4.5 HMMT and IFEval); the base model maker's own measurement (Genie Cortex SWE-bench Pro v2); lmmarketcap (GPT-5 IFEval).
Declared sources: OpenAI's GPT-5 for developers announcement; Anthropic's Claude Opus 4.5 announcement and system cards; the base models' technical reports and model cards.
Get one for your website:
Choose a product and leave your email and phone. Our sales team will get back to you.
Your email is confirmed. Our sales team will get back to you shortly about .
Tell us a little about what you need. Our team will get back to you, usually the same day.
Your email is confirmed. Our sales team will get back to you shortly about .