Davies Meyer – home
    Tech3 min read

    NLWeb

    NLWeb is an open project offering protocols and tools for natural-language website queries. It uses structured content and supports MCP so people and connected AI agents can ask a website questions. NLWeb is not general software for visually operating arbitrary websites, nor does it grant automatic purchasing authority.

    NLWeb explained

    A practical starting point is a question about a known collection: which products meet certain criteria, or which event fits a date range? The reference implementation processes structured and returns structured answers. It therefore uses available data and interfaces rather than fundamentally replacing them.

    Useful deployment needs current, unambiguous and accessible content. Define which collections can be queried and how changes are incorporated. An answer can sound plausible while missing a limitation. Check statements against the underlying records and provide a reliable route to the source.

    A natural-language query does not automatically create a booking, purchase or CRM update. Such actions need separate integrations and permissions. Privacy, access protection, resistance to manipulative inputs and ongoing operating costs also remain implementation tasks. The project does not universally promise less maintenance.

    Creative Engineering starts by asking which orientation problem the dialogue should improve. We take responsibility for the concept and quality. Compare a limited prototype with existing search: do people find suitable information, understand the answer and know what to do next? Adding a chat interface is not progress by itself.

    Examples

    Hypothetical application

    An organiser makes an approved event catalogue queryable. Users can ask by date and event type. The dialogue links suitable entries and acknowledges missing details; actual booking happens in the designated system.

    Key Points

    • Anchor language queries in a defined content collection.
    • Structured data and interfaces remain important.
    • Check answer quality and the route to the next step.

    Practical application

    Choose a maintained catalogue and specific questions. Test a limited dialogue against existing search, including incorrect and unanswerable requests.

    Useful measures

    Factually correct answers

    Check answers and their limitations against approved source data.

    Successful orientation

    Measure whether users find the suitable record and understand the next step.

    Total effort per resolved query

    Include data maintenance, model operation, monitoring and necessary rework.

    Common mistakes

    • Equating NLWeb with universal visual browser automation.
    • Covering missing data with invented answers.
    • Publishing an interface without updates, access protection and an operating plan.

    Sources and context

    Frequently Asked Questions about NLWeb

    No. The project provides protocols and tools for language-based queries. Visually clicking through arbitrary websites is a different capability.

    Loading related terms…

    All Terms