Semantic Definition Format (SDF) for API translation

Introduction In many cases, a service needed by an application can be provided through APIs by several suppliers. Semantic matching can be used to map the needs of the applications and adapt API calls to what the service suppliers could provide. The main purpose of this document is to describe methods that enable the translation of one device or application API to another equivalent API, utilizing syntax understandable by the deploying party. An API translator from the original API to the updated one or even mapping to a new service would allow an automatic adaptation to the target API. Also, it would enable the possibility to adapt to multiple APIs with the same data model, enabling the deployment of services from multiple sources. This feature enables use cases such as the usage of the cheapest or most robust service, deployment of national services when traveling, and execution of device services even from a different ecosystem origin. This document provides enabling elements for communication from devices or applications belonging to different ecosystems that do not share the same semantic modelling. Utilizing SDF syntax enhanced with an API format depiction, it is possible to provide an API description that is understandable with the syntax known by the device or application, if such semantics are defined in their data model. The final result uses a navigable description of the API's functions to provide the necessary information that is required by a device application to map external required services, their parameters, and results to an application logic using ontologically defined objects that can be matched to remote calls in another device application, for example. The API requires a mapping in applications functions by matching syntax in addition to the actual discovery of the API; therefore, the importance of the translation to an understandable model. This matching could be enabled by several mechanisms described, and it is dynamically adjusted to every new interface faced. Using SDF in conjunction with the given schema allows an automated way to match functionality to application intents. Example of a device using the API translation model to access the same service in a different platform with a different API

Terminology The definitions of apply. The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in . As explained in the words apply when, and only when, they appear in all capitals, as shown here.

Discoverable API interfaces A device or application may expose its capabilities as an API list. This could be acquired in real-time by calling a command that provides such a list of commands. For example, the resource “.well-known/commands” could be used as a specific way to request all the commands available. The reply then includes a list of the different API supported calls, which would be called in a hierarchical way from the root domain that is provided by the device, an application, or a mediating platform. The reply would contain a description of the supported API with a semantic definition attached. Using the Semantic Definition Format (SDF), a reply for two commands may include what is given in the following example: In this example, one service is defined with a command1, which requires an input1 (an object) and input2 (a number). The other service requires a different call (command2), which requires only one input (secondInput). Further interaction capabilities can be included either in the initial response, as shown above in the "sdfAction" block, or could be discovered by further requests. For example, the descriptions of the two actions detailed above show if they require input, what the details of such input and the output are, and if there is additional information related to the nature of the output. Also, the description field could be tailor-made to provide a URI to a vocabulary that is understandable by the device or application. In this way, the device or application could make sense of the possible options, parameters, and actions that the command provides or requires.

Inputting parameters to the API Once the available function calls and input parameters required by the API have been provisioned to the requester device or application, it can make calls to the API when required by its application logic. The application would map the “need for a capability” to the concrete capability provided by the provider application or device through matching the syntax and data model. The functions are then mapped to the calls that the application can handle and has defined syntax. The parts of the API that are not understood by the requester device are normally ignored. The mapping semantics from a target device or application may be communicated to the application of the requesting device using an SDF mapping extension draft . In such mapping, the actions and data can be expressed using the data model that are understandable by the requester and allows it to produce the API call and provide the required data. The only problem with this mapping is that if the API call structure is different, the logic required to call the target API might be complicated for certain device applications. With SDF notation, it is possible to explain the data model but not the mapping for a specific API call. That would require, in some cases, mapping of values and, depending on the API definition, reshuffling of the parameters or even the URI utilized. To tackle those challenges, we introduced a new SDF quality that can express URL template patterns from . The quality "template-href", includes the URL or the URL pattern that matches the command known by the requester device application. As explained in , the pattern may have variable terms, which are also given an SDF definition and relationship so the requested device can understand how to map its own data model to what the target API requires. Also, we introduce a value mapper "template-value" that enables mapping of values to parameters in the target API. The sdfRelation is used as a vehicle to support the mapping between APIs. The namespace "cmd" MUST be utilized to provide this API mapping values. In this example, we map a command_1 with several inputs to a command_2 that utilizes the inputs in a different way. The schema maps semantically the inputs and instructs the system how to make the API call based on the previously known API.

Methods on the API calls Sometimes, only calling an API is not enough. For example, in some cases, the API call requires additional actions such as submitting a payload. To handle this dimension on the API calls, the method command has been introduced. This command SHALL enable the requested device to understand how to handle the API call, since in the previous example, the API URL has all the information required in the call itself. However, in some cases, the action necessary with the API may need to be specified in a manner that the requested device can understand. The keyword method is meant to describe how MUST be handled the API URL and what to expect of it. In some other examples, the method could result in an OBSERVE or another type of new action not yet specified, for example, how to address an XR-type application. An example could be the use of an LWM2M camera . To use this model, a payload has to be submitted in the POST method. The following example describes such a translation from one camera API to the LWM2M model.

Matching API to application logic Each application has a logic that requires either information from the API call or the execution of a service. With this translation tool, the application logic may need to consider the different types of APIs that may be present. For example, if the API requires a payload or the info is directly appended to the call. Also, on the way the API produces the output, and how to handle it.

Agentic AI Integration AI agents rely heavily on natural language to understand APIs. But mistakes can be made without a full understanding of semantics. With the help of SDF, the semantics can be clarified and used as intended. Instead of forcing the LLM to output the translated raw payload token-by-token, an API translation could be utilized to indicate a new API deployment. Nowadays, MCP (Model Context Protocol) provides a standard way for AI agents to discover APIs descriptions, invoke translated operations safely, and receive structured results. SDF definitions and SDF API translation descriptions may be exposed to agentic systems as MCP resources. For example: - sdf://devices/thermostat-123/model. The result of the calls can be an SDF model. - sdf://devices/thermostat-123/translation. After providing the SDF model of the device, the system provides a translation on how to use the service provided. - sdf://translations/vendor-x-to-vendor-y. Provides a translation from a system from vendor X to a system from vendor Y.

Ontologies describing APIs Every application domain has an ontology or vocabulary that indicates actions and their parameters. These actions can be translated to APIs but they do not necessarily explain the API itself. Just the result of the API action. Some ontologies describing APIs may need to be involved in some cases. Such as W3C Web of Things (WoT) Thing Description (TD), Hydra Core Vocabulary (for hypermedia Web APIs) and others with less focus in APIs but relevant like OWL S / WSMO (Semantic Web Services) , schema.org Actions and OGC SensorThings API .

IANA Considerations This document has the following actions for IANA. Note to RFC Editor: Please replace all occurrences of "{XXXX}" with the RFC number of this specification and delete this paragraph.

Media Type IANA is requested to add the following Media-Type to the "Media Types registry . --> A media type for SDF mapping files

Name	Template	Reference
sdf-mapping+json	application/ sdf-mapping+json	RFC XXXX, section 5.1

Security Considerations Some wider issues are discussed in .

Acknowledgements Thanks to Ari Keränen for the support in realizing this idea.