Skip to main content
    Service 01

    AI Chatbot & RAG Assistant

    Not a scripted bot. An assistant that reads your knowledge base and past tickets before it answers, cites what it used, books into a real calendar and escalates to a named person when it is unsure. Runs on your API keys, in your accounts.

    Price
    from $4,000
    Time to live
    2 weeks
    Price it in the calculator ›
    AI Chatbot & RAG Assistant

    What it does

    • 01Vector store of your docs, policies and past tickets, refreshed on a schedule you set
    • 02Citation-grounded answers: no invented prices, no invented policies
    • 03Tool use: book calls, check order status, open a ticket, escalate
    • 04Web widget, Slack, Teams, Telegram and WhatsApp from one assistant
    • 05Guardrails your team edits: topics it must not touch, tone, languages
    • 06Evaluation set and answer-accuracy report before launch and every month after

    What ships in the first two weeks

    1. Week 1: knowledge base indexed, assistant answering the evaluation set with citations
    2. Week 2: live on the first channel with escalation to a named person and analytics
    3. Handover: runbook, evaluation set, three months of fixes

    Every deliverable ends with code in your repository, a runbook and three months of fixes. Scope and price are fixed in writing before we start.

    Questions

    01Will it make things up?

    It answers only from the documents we index and shows the source next to the answer. When the retrieved text does not cover the question it says so and routes to a person instead of guessing. We measure this on a fixed evaluation set before launch.

    02Where does our data go?

    Into your own vector database and your own model account. We build in your cloud or on your hardware; nothing is routed through us. For regulated data we deploy an on-premise model instead of a public API.

    03Which model does it use?

    Whichever fits the task and the budget: we price the same task across Claude, GPT and Gemini in the cost calculator and pick with you. The assistant is built so the model can be swapped later without rewriting it.

    04How is success measured?

    Share of conversations resolved without a person, answer accuracy on the evaluation set and time to first response. All three are baselined before launch and reported monthly.

    Go deeperAI for customer support: a practitioner's playbook

    Scope this in a 30-minute call

    You describe the process, we tell you what a first version costs and how long it takes. Free, no slides.

    ‹ All services