A private AI assistant for your business
An assistant that answers in WhatsApp, Telegram, Slack or Teams and runs on a server you control — not inside somebody else's product. Your correspondence and documents stay where you put them.
Your team keeps writing in the chats they already use. The gateway runs on your own machine, and the only thing that leaves it is the request you allowed, to the model account you hold.
What it actually is
OpenClaw is an open-source, self-hosted gateway. It sits between the messaging apps your team already uses and an AI model of your choice, and it runs on your own machine — either a managed server from us or a small device in your own office.
Your staff carry on writing in the same chat they use today. There is no new application to learn and no third-party product holding your conversations.
What it does day to day
- Drafts replies to routine enquiries, in your tone, for a person to send
- Summarises long threads, documents and voice notes
- Sorts incoming requests and flags the ones that matter
- Remembers what was promised and reminds the right person
- Looks things up in the files and systems you give it access to
The work it saves is the work nobody enjoys: reading, sorting and repeating.
What we do not sell
An autonomous “AI employee”. This is a tool that drafts, summarises, triages and reminds, under defined access and human approval. Anything that sends, pays, publishes or deletes waits for a person to confirm it.
We would rather set that expectation now than after the invoice.
Deployed safely, or not deployed at all
An agent with access to a terminal and a filesystem is exactly as dangerous as it sounds if it is installed casually. Every deployment we do runs on an isolated machine with no other business data on it, behind a firewall, reachable only over key-based access.
The assistant answers only the numbers you allowlist, on a dedicated number rather than your personal one, and spending on the AI model is capped before it ever runs.
Where it runs
Our managed AI assistant server
£95 per month. Eight vCores, 16 GB of memory and 480 GB of NVMe storage, hardened, monitored and backed up by us. Nothing for you to look after.
Your own Umbrel Home
From £480, once. The same 16 GB specification on a palm-sized device that lives on your own network. No monthly hosting fee from us at all.
We will tell you which of the two suits you before you buy anything.
What it costs
- Deployment, remote — £750. Installation and hardening, two channels connected, model selection with spending limits, instructions written for your business, one hour of training and fourteen days of support.
- Deployment, on site in London — £1,650. Everything above, installed at your office, with half a day of training for the team and thirty days of support.
- Ongoing care — £150 per month. Updates, monitoring of every connected channel, backup checks, a monthly review of what the model actually cost, and up to two hours of changes.
- The AI model — your own account with Anthropic or OpenAI, typically £5–30 per month for a small team. You hold the key and see the bill.
- A separate number, only if you want the assistant in WhatsApp. WhatsApp ties an account to a phone number, so the assistant needs its own rather than sharing yours. A spare pay-as-you-go SIM is enough, and an existing spare business number works too. Telegram, Slack and Teams need no number at all.
Remote work and an on-site build are genuinely different jobs, and both prices are published above so you can decide which one you actually need. If a hosted assistant in somebody else’s cloud is enough for you, that is cheaper still — around £50 a month — and we will say so rather than sell you a deployment.
How deployment goes
Step 1
A short call
What you want the assistant to do, and what it must never touch.
Step 2
Server and hardening
Isolated machine, firewall, key-based access, spending limits.
Step 3
Channels and instructions
Two channels connected, the workspace and instructions written for your business.
Step 4
Training and handover
An hour remotely or half a day on site, then two to four weeks of support.
Who it suits
It works well for teams answering the same questions every day, for businesses that want the data on hardware they own, and for owners who spend evenings catching up on messages.
It is the wrong purchase if you expect it to act without supervision, if nobody in the company will own its instructions, or if your work needs judgement we cannot write down.
Related
The assistant sits on a server we look after — see server support. Security of that machine is covered on cyber security and backup. Deployment, monthly care and the server itself are in our shop.
Tell us what takes your time
Book a free thirty-minute consultation. We will look at how your team actually works and say plainly whether an assistant is worth it for you.