Every client is asking for an AI assistant. Now you can say yes, and prove it works.
Finchley turns a client’s documents into a support assistant you can test, show them, and put on their site with one line. Then charge for it every month, on your own model keys, with nothing marked up.
“Can we have a chatbot?” is now on every brief. Saying yes has meant a one-off build each time.
A developer can wire up an API in a week. Then prompts drift, documents change, keys expire, and nobody on the account team can fix it.
The first time anyone finds out the bot says “14 days” instead of “30” is a customer, on the client’s site.
Three steps. Under five minutes to the first useful answer.
Drop the client’s documents, a folder, or their website. Finchley reads them straight away and keeps websites fresh every day. Scanned menus and photos are read too.
Share a readiness page the client can open across a table: the pass rate, what it gets right, and what still needs a source. Behind it, Finchley proposes test questions from the documents and grades every answer.
Paste one script tag. The widget picks up the name, colours and knowledge from Finchley, so there is nothing else to change on the client’s site, ever.
Everything an agency needs to run assistants for clients, in one quiet place.
PDF, Word, Excel, PowerPoint, CSV, email, images, whole folders, and websites checked daily. Replace a file and the assistant knows straight away.
A shareable page with the pass rate and what still needs sorting, in plain words. Open it in the pitch, send it before launch, and let the client sign off on evidence.
Connect Shopify, HubSpot or any MCP server. Every action is Off, Allowed, or Ask first. Anything that writes asks the customer before it acts.
Every real conversation, with citations and tool calls. Flag an answer, add a note, and turn it into a test so it can’t slip again.
Your colours on the dashboard and the widget, your own domain for sign-in and client pages, and no “Powered by” on Studio.
Model usage runs on your own Anthropic and OpenAI keys. Finchley never marks up tokens, and each client has a spend limit you set.
It never speaks. It just tells you how things are.
A small mark lives next to each client and inside every widget. It reads while documents come in, thinks while it answers, looks up when a test fails and closes its eyes when everything passes. Agencies can hide it or use the client’s logo instead.
One flat monthly price. Your model keys, so there is nothing to mark up.
- Up to 3 client workspaces
- Knowledge, tests, tools, logs and the widget
- Your accent colour on the widget
- “Powered by Finchley” on the widget
- 14 days free, no card
- Unlimited client workspaces
- Knowledge, tests, tools, logs and the widget
- White-label colours and your own domain
- No “Powered by” on the widget, client viewer seats
- 14 days free, no card
Studio is £199 a month for unlimited clients. Charge each client £150 a month for a managed assistant and 10 clients is £1,301 a month in margin, before model usage on your own keys. The subscription is flat, so every client after the first is almost all margin.
Model usage runs on your own Anthropic and OpenAI keys and is billed by them at their rates. Finchley never marks up tokens. Prices exclude VAT. The margin example uses a per-client price agencies commonly charge; yours is up to you.
Things agencies ask before they start.
Why do I bring my own model keys?
So your costs are your costs. You pay Anthropic and OpenAI directly at their rates and set a monthly spend limit per client. Finchley is a flat subscription and never adds a margin to tokens.
What should I charge my clients?
Most agencies charge between £150 and £300 a month per client for a managed assistant, sometimes bundled into a retainer. The Finchley subscription is flat, so beyond the first client or two the rest is margin. Model usage is on your keys and is usually a few pounds a month per client.
Do I need a developer?
No. Documents in, tests, one script tag on the client’s site. A developer is only needed if you want to connect a custom MCP server or write your own tools.
What happens to a client’s documents?
They are read into a private index for that workspace and used only to answer that client’s customers. Nothing is used to train anyone’s models. Remove a workspace and everything is kept for 30 days, then deleted.
What files can it read?
PDF, Word, Excel, PowerPoint, Markdown, CSV, plain text, email exports, images and zips or whole folders. Websites are fetched and checked for changes every day. Scanned PDFs and photos are transcribed.
How do the tests work?
A test is a question and the behaviour you expect, in plain words. Finchley asks the assistant and a second model grades the answer against your expectation, then explains any miss. Finchley proposes tests from the documents; you keep the ones you agree with.
Can my client see it?
Yes. Share a read-only readiness page that updates every time tests run, and invite people at the client to see logs and tests for their workspace only.
Can I cancel?
Any time from the billing page. The assistant keeps answering until the end of the period you have paid for.
Running more than a handful of clients? Twenty minutes with us and you’ll know if it fits.
Bring a real client’s documents. We will set the assistant up with you on the call, run the first tests, and talk through pricing for your book of clients.
Early access. We onboard the first 20 agencies personally.
Ship the first client this week.
Add a client, drop their documents, and have a tested assistant on their site before the next status call. Then send the invoice.
Start free trial