Filter and merge request body fields

Use the filterKeys() and merge() CEL functions together with json() and toJson() to sanitize a JSON request body before it reaches the upstream. filterKeys() removes unwanted fields by testing each key against a predicate. merge() combines two maps, with the second map’s value overwriting any matching keys in the first. toJson() serializes the resulting map back to a JSON string.

Before you begin

1. Set up an agentgateway proxy.
2. Install the httpbin sample app.

Filter and merge request body fields

In this example, an incoming request body contains internal metadata fields prefixed with x_ that should not be forwarded to the upstream. After stripping those fields, default values for model and max_tokens are merged in.

For example, a request body of {"messages": [...], "model": "gpt-3.5-turbo", "x_trace_id": "abc", "x_user_session": "xyz"} is forwarded upstream as {"messages": [...], "model": "gpt-4o", "max_tokens": 2048}.

1. Create an AgentgatewayPolicy resource with your transformation rules.

   kubectl apply -f- <<EOF
   apiVersion: agentgateway.dev/v1alpha1
   kind: AgentgatewayPolicy
   metadata:
     name: transformation
     namespace: httpbin
   spec:
     targetRefs:
     - group: gateway.networking.k8s.io
       kind: HTTPRoute
       name: httpbin
     traffic:
       transformation:
         request:
           body: 'toJson(json(request.body).filterKeys(k, !k.startsWith("x_")).merge({"model": "gpt-4o", "max_tokens": 2048}))'
   EOF

   The expression breaks down as follows:

   • json(request.body): Parses the raw request body string into a map.
   • .filterKeys(k, !k.startsWith("x_")): Keeps only the keys that do not start with x_, stripping internal metadata fields.
   • .merge({"model": "gpt-4o", "max_tokens": 2048}): Merges in the default values, overwriting any existing model value and adding max_tokens.
   • toJson(...): Serializes the resulting map back to a JSON string for the request body.

2. Send a POST request to the httpbin app with a JSON body that includes internal metadata fields. Verify that you get back a 200 HTTP response code and that the x_ fields are absent from the forwarded body.

   Cloud Provider LoadBalancerPort-forward for local testing

   curl -vi http://$INGRESS_GW_ADDRESS:80/post \
    -H "host: www.example.com:80" \
    -H "content-type: application/json" \
    -d '{"messages": [{"role": "user", "content": "hello"}], "model": "gpt-3.5-turbo", "x_trace_id": "abc123", "x_user_session": "xyz789"}'

   curl -vi localhost:8080/post \
   -H "host: www.example.com" \
   -H "content-type: application/json" \
   -d '{"messages": [{"role": "user", "content": "hello"}], "model": "gpt-3.5-turbo", "x_trace_id": "abc123", "x_user_session": "xyz789"}'

   Example output:

   < HTTP/1.1 200 OK
   HTTP/1.1 200 OK
   < content-type: application/json
   content-type: application/json
   ...
   
   {
     "data": "{\"messages\":[{\"role\":\"user\",\"content\":\"hello\"}],\"model\":\"gpt-4o\",\"max_tokens\":2048}",
     ...
   }
   

   The x_trace_id and x_user_session fields are absent. The model value is overwritten to gpt-4o and max_tokens is added.

Cleanup

kubectl delete AgentgatewayPolicy transformation -n httpbin