diff --git a/generic-data-exchange-api.md b/generic-data-exchange-api.md
new file mode 100644
index 0000000..355ad20
--- /dev/null
+++ b/generic-data-exchange-api.md
@@ -0,0 +1,243 @@
+# Generic Data Exchange API
+
+This API specification defines a protocol for how a client can consume data from a server and update a local database or store; or push data to a remote endpoint and modify a collection of resources.
+
+A compliant server exposes a number of datasets. Each dataset contains resources of one or more type and exposes a feed of changes. Each change is represented as the latest representation of a resource that has been created, modified or deleted. Each resource has unique identity across all resources in all datasets exposed by a single server.
+
+## Motivation
+
+The main motivation of this API is to enable the synchronisation of ICAR resources between compliant services in a generalised manner and without clients needing to be location aware. 
+
+As the API is generic and not bound by resource type this API enables the addition and exchange of new ICAR resource types without need for API revision. 
+
+As the API supports both PULL and PUSH semantics it can be used in a composable way to meet many use case, e.g. from farm to aggregator (push) and between aggregators (pull).
+
+The API is designed to be super simple and bring a low bar of entry to any software to enable it to participate in a data sharing eco-system. There are just 4 endpoints defined in this API yet that is enough to enable the exchange of any form of ICAR resource.
+
+## Concepts
+
+# Protocol Definition
+
+The HTTP(s) API has a single entry point and is consistently structured to allow introspection and dynamic navigation. A compliant server exposes the following endpoints.
+
+## Datasets Endpoint
+
+The datasets endpoint returns a list of datasets that it is exposing. 
+
+`/datasets` 
+
+The following Schema is used to represent a single dataset in the list:
+
+```json
+{
+    "description": "Dataset Resource.",
+    
+    "type": "object",
+    
+    "required": [
+        "name", "url", "changes"
+    ],
+
+    "properties": {
+        "name": {
+            "type": "string",
+            "description": "Unique dataset name exposes by this service endpoint."
+        },
+        "url": {
+            "type": "string",
+            "description": "Dataset url"
+        },
+        "changes": {
+            "type": "string",
+            "description": "Url that exposes the changes for this dataset"
+        },
+        "containedTypes": {
+            "type": "array",
+            "description": "An array of strings. Each value is the name of a resource type exposed in this dataset.",
+            "items" : {
+                "type" : "string"
+            }
+        }
+    }
+}
+```
+
+An example response from a compliant endpoint could look like the following:
+
+```json
+[
+    {
+        "name" : "animals",
+        "url" : "/datasets/animals",
+        "changes" : "/datasets/animals/changes",
+        "containedTypes" : [ "icarAnimalCoreResource"]
+    },
+    {
+        "name" : "everything",
+        "url" : "/datasets/everything",
+        "changes" : "/datasets/everything/changes"
+    }
+]
+```
+
+## Dataset Endpoint
+
+The dataset endpoint allows a client to retrieve a single dataset. 
+
+`/datasets/{datasetid}`
+
+The schema of the dataset resource is the same as the dataset resource when returned in the list. An example response is shown below.
+
+```json
+{
+    "name" : "animals",
+    "url" : "/datasets/animal",
+    "changes" : "/datasets/animals/changes",
+    "containsTypes" : [ "icarAnimalCoreResource"]
+}
+```
+
+## Changes Endpoint
+
+The changes endpoint named in the dataset resource exposes a changes feed of resources. The response from this request is an array of JSON objects. The first object is a context that is reserved for future use. The last object is a continuation object that allows clients to consume subsequent changes. 
+
+`/datasets/{datasetid}/changes?since={sinceToken}`
+
+The changes endpoint returns the following JSON response:
+
+```json
+[
+  {
+     "id" : "@context",
+     "description" : "application specific / expansion point for rdf, JSON-LD and Entity Graph Data Model" 
+  },
+
+  {
+      "resourceType" : "icarAnimalCoreResource",
+      "location" : {
+          "id" : "a-location-id",
+          "scheme" : "io.mimiro.data.locations"
+      },
+      "meta" : {
+          "source" : "io.mimiro.data.cattle",
+          "sourceId" : "some-cow-id-1"
+      }
+  },
+
+  {
+      "resourceType" : "icarAnimalCoreResource",
+      "location" : {
+          "id" : "a-location-id",
+          "scheme" : "io.mimiro.data.locations"
+      },
+      "meta" : {
+          "source" : "io.mimiro.data.cattle",
+          "sourceId" : "some-cow-id-2",
+          "isDeleted" : true
+      }
+  },
+
+  {
+      "id" : "@continuation",
+      "token" : "token-issued-by-server-to-get-more-data"
+  }
+]
+```
+
+The first object in the response is for future use and will contain context information. Likely uses are for JSON-LD, or the Entity Graph Data Model.
+
+The main set of objects in the response are ICAR resources. To be valid in the context of this protocol several properties are required:
+
+`location` : Must be specified as this is the main logical grouping of resources when being exchanged.
+
+`meta/sourceId` : The source id of the meta structure must be present and as stated, it must be a unique id related to the source server. e.g. Values of this must be unique across all resources.
+
+`meta/source` : The source of the meta structure must be present.
+
+`resourceType` : The value of this property must be a valid ICAR resource schema, e.g. icarAnimalCoreResource. This is also the requried discriminator property.  
+
+When a client receives the ICAR resources it must process them in order. For each entity it MUST do the following: 
+
+1. check if the `meta/isDeleted` flag is set to `true`. If so the local resource with the corresponding sourceId MUST be deleted or marked as deleted depending on the local system semantics.
+
+2. If there is no local version of the resource it MUST be created and the sourceId must be associated with the local object in some way. 
+
+3. If there is a local object with the correlating sourceId then this representation must be deleted and replaced with the one received. Implementations are free to perform deltas, but the result must be the same. 
+
+
+The last object in the array is a continuation object and contains a property `token`. The token is an opaque string encoded with base64. This token is issued by the server and clients should NOT manipulate the contents. Manipulating the contents of a continuation token gives undefined behaviour when used as part of a subsequent request. 
+
+A client is responsible for storing the continuation token and then using it as the since token in subsequent calls. e.g.
+
+`/datasets/{datasetid}/changes?since=token-issued-by-server-to-get-more-data`
+
+Note that the continuation token can be used by the server to provide paging. A client typically calls to the changes endpoint until no resources are returned. It then stores the token and then waits for some period before asking again. 
+
+If a server has reloaded all data in a dataset and needs the client to resync from the beginning, then the server can return an HTTP Header:
+
+`icar-full-sync: true`
+
+If this is returned the client should delete all local data related to this dataset, discard any continuation tokens, then make a call to `/changes` and continue as above.
+
+The client should only store the contiuation token if it has succesfully processed all resources returned in a response.
+
+## Dataset Push 
+
+The datasets concept can also be used in a push based scenario. In this model a client POSTs resources to the dataset's `resources` endpoint. 
+
+POST `/dataset/{datasetid}/resources`
+
+```json
+[
+  {
+     "id" : "@context",
+     "description" : "application specific / expansion point for rdf, JSON-LD and Entity Graph Data Model" 
+  },
+
+  {
+      "resourceType" : "icarAnimalCoreResource",
+      "location" : {
+          "id" : "a-location-id",
+          "scheme" : "io.mimiro.data.locations"
+      },
+      "meta" : {
+          "source" : "io.mimiro.data.cattle",
+          "sourceId" : "some-cow-id-1"
+      }
+  },
+
+  {
+      "resourceType" : "icarAnimalCoreResource",
+      "location" : {
+          "id" : "a-location-id",
+          "scheme" : "io.mimiro.data.locations"
+      },
+      "meta" : {
+          "source" : "io.mimiro.data.cattle",
+          "sourceId" : "some-cow-id-2",
+          "isDeleted" : true
+      }
+  }
+]
+```
+
+The semantics for PUSH are the same as for the pull semantics described above with the following exceptions:
+
+1. Full sync is not accepted as header in the push scenario. 
+2. The payload for push MUST include only a context followed by resources. i.e no continuation token json object.
+
+The server returns a 200 OK if the new resource representations are correctly processed. 
+
+Any other response code and the client MUST assume that the server has not received and stored the latest state. 
+
+Use cases for push is typically small on site devices that are pushing data up to a cloud location which then collects resources from many clients before sharing them further or processing them. It is best practice to have different dataset endpoints for Push and Pull. 
+
+# Security
+
+All communication SHOULD be over HTTPs.
+
+It is Recommended to use JWT tokens to secure API endpoints. 
+
+The server is responsible for mapping a client to a set of claims which give access to certain locations. The set of resources exposed MUST only be for the locations that a given client can access. 
+
+The list of datasets exposed to a client should also be limited by client ACL or claims. 
\ No newline at end of file