intent-runtime

    Introduction

    Intent Runtime turns user intent into a structure your application or MCP tools can read. Give it a request and relevant material; receive goals, requirements, prohibitions, business parameters, and questions that need clarification.

    For “Look up order 000123. Do not cancel the order,” it preserves the lookup goal and the prohibition. With an orderId Schema, it can also preserve the string "000123". Your application decides whether to call its order system. Intent Runtime does not query or cancel orders.

    When to use it

    Use it before handing natural-language requests to application logic, when you need to preserve goals, prohibitions, conditions, and unresolved questions. Configure business fields only when you need parameters such as an order ID. It is not an order system, business task manager, or autonomous execution agent.

    Input and output

    You supplyYou receive
    Original inputGoals and constraints expressed by the user
    Free-form JSON context or an optional baselineSupported references, additions, changes, and withdrawals
    Optional business Schema and fieldsParameters with explicit ownership, validation status, and issues
    Optional business semanticsActions and objects interpreted using your domain vocabulary

    Intent recognition works without a Schema. Field extraction is optional. Understanding, field validation, and business authorization are separate layers; see Read the result.

    “Analyze this project” can express an analysis goal. If no project material is supplied, the component does not read a repository, invent project capabilities, or create an implementation plan. Unresolved references remain questions.

    Choose an integration, run the quick start, then learn to read the result. Full parameter and type definitions are in the reference; you do not need to read them all first.

    It understands only caller-supplied material. Goals and ordering come from the user's expression. It does not search projects or history, choose business tools, generate implementation plans, manage business tasks, authorize, or execute operations. Model generation and local validation are dependencies of recognition.

    The application owns messages, model selection, results, and business state. Supply relevant context or the previous effective snapshot explicitly for subsequent turns. Model candidates undergo source, business-rule, identity, and public-structure checks. Passing programmatic checks is not a natural-language accuracy guarantee.

    Next step

    Choose an integration, or make your first Node.js call.

    The npm package version and result format version are separate identifiers. Successful results and baselines use schemaVersion 2.0-draft.6.