AI Chatbots and Knowledge Bases
We build AI assistants that answer questions from your own documents and systems, on your website or for your staff. Every answer cites its source, questions the documents do not cover are declined or handed to a person, and the assistant is tested against real questions before it goes live.
What we build
- Website assistants that answer customer questions about your services, policies, pricing ranges, and process, and hand off to a person or a booking form when that is the next step.
- Internal knowledge assistants that answer staff questions from manuals, policies, past proposals, and help articles, with a link to the source.
- Assistants that act: looking up an order or job status, opening a ticket, or starting a workflow, with approval steps where the action matters.
How it works
The assistant uses retrieval-augmented generation (RAG): it finds the passages in your documents that relate to a question and has a language model answer from those passages only, citing them. Your documents are not used to train a model, and updating an answer means updating the document. The full explanation is in our guide to building an AI knowledge base.
What makes it trustworthy
- Citations on every answer, so people can check the source.
- Declines instead of guessing when the documents do not cover a question, and offers a person instead.
- A test set of real questions with known answers, run before launch and after every change.
- Logs of every conversation, so you can see what people ask and where the documents fall short.
- Permissions, so staff and clients see only the documents they should.
When off-the-shelf is enough
For a small set of documents used internally, the business plans of ChatGPT, Claude, and Microsoft 365 Copilot can already answer from uploaded or shared files. We will say so if that covers your need. A custom build makes sense for customer-facing answers, many sources kept current, per-person permissions, or actions in your own systems.
Technical approach
- Models from providers such as Anthropic and OpenAI, under commercial API terms that exclude training on your data.
- Documents indexed from where they already live, such as a shared drive, a help center, or a database, and re-indexed as they change.
- Hosting, data, and API keys in accounts in your name.
Frequently asked questions
- Can an AI chatbot on our website give wrong answers?
- It can, which is why it answers only from your documents, cites them, declines what they do not cover, and is tested against real questions before launch and after every change.
- Do we need to train a model on our documents?
- No. Retrieval keeps your documents separate from the model, which is cheaper, easier to keep current, and lets every answer cite its source.
- What does an AI chatbot cost to run?
- Mostly model usage, billed per question by the provider, plus hosting. We estimate the cost per conversation before building. The build itself is quoted per project or hourly; see pricing.
- Can it hand a conversation to a person?
- Yes. It can collect the details and send them to your team by email or chat, or offer a booking link, whenever a question needs a person.