feat: honor Doubleword model service tiers (#4)

Author: hcastc00

apps/gateway/internal/adapters/http/mapper/chat_mapper_test.go

@@ -14,7 +14,7 @@ func read(t *testing.T, body string) (domain.GenerateRequest, error) {
 }
 
 func TestChatCompletionRequestToDomain_ReadsTheSummary(t *testing.T) {
-	req, err := read(t, `{"model":"m","stream":true,"max_completion_tokens":50,"parallel_tool_calls":false,
+	req, err := read(t, `{"model":"m","stream":true,"max_completion_tokens":50,"service_tier":"flex","parallel_tool_calls":false,
 		"response_format":{"type":"json_object"},"temperature":0.2,"unknown_extension":{"x":1},
 		"messages":[{"role":"developer","content":"be brief"},
 			{"role":"user","content":[{"type":"text","text":"What is "},{"type":"image_url","image_url":{"url":"data:image/png;base64,AA"}},{"type":"text","text":"this?"}]},
@@ -24,7 +24,7 @@ func TestChatCompletionRequestToDomain_ReadsTheSummary(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if req.PublicModelID != "m" || !req.Stream || *req.MaxOutputTokens != 50 || *req.ParallelToolCalls || !req.StructuredOutput {
+	if req.PublicModelID != "m" || !req.Stream || *req.MaxOutputTokens != 50 || req.ServiceTier == nil || *req.ServiceTier != "flex" || *req.ParallelToolCalls || !req.StructuredOutput {
 		t.Fatalf("summary %+v", req)
 	}
 	if len(req.Input) != 4 || *req.Input[0].Role != "developer" || *req.Input[1].Content != "What is this?" || req.Input[2].Content != nil || *req.Input[3].Content != "42" {

apps/gateway/internal/adapters/http/mapper/chat_to_domain.go

@@ -19,6 +19,7 @@ type chatRequest struct {
 	Stream              *bool           `json:"stream"`
 	MaxTokens           *int            `json:"max_tokens"`
 	MaxCompletionTokens *int            `json:"max_completion_tokens"`
+	ServiceTier         *string         `json:"service_tier"`
 	N                   *int            `json:"n"`
 	Tools               []chatTool      `json:"tools"`
 	ToolChoice          json.RawMessage `json:"tool_choice"`
@@ -77,6 +78,7 @@ func ChatCompletionRequestToDomain(raw json.RawMessage) (domain.GenerateRequest,
 		PublicModelID:     req.Model,
 		Stream:            req.Stream != nil && *req.Stream,
 		MaxOutputTokens:   req.MaxTokens,
+		ServiceTier:       req.ServiceTier,
 		ParallelToolCalls: req.ParallelToolCalls,
 		StructuredOutput:  req.ResponseFormat != nil && (req.ResponseFormat.Type == "json_object" || req.ResponseFormat.Type == "json_schema"),
 	}

apps/gateway/internal/adapters/metering/runtime_client.go

@@ -330,6 +330,7 @@ func (m runtimePublicModel) toDomain() domain.PublicModel {
 		Description:                   m.Description,
 		ProviderModelID:               m.ProviderModelID,
 		UpstreamModelName:             m.UpstreamModelName,
+		ServiceTier:                   m.ServiceTier,
 		ProviderConfig:                m.ProviderConfig.toDomain(),
 		SupportsChatCompletions:       *m.SupportsChatCompletions,
 		SupportsChatCompletionsStream: *m.SupportsChatCompletionsStream,
@@ -349,6 +350,7 @@ func (m runtimePublicModel) toDomain() domain.PublicModel {
 		model.Fallback = &domain.ProviderTarget{
 			ProviderModelID:   m.Fallback.ProviderModelID,
 			UpstreamModelName: m.Fallback.UpstreamModelName,
+			ServiceTier:       m.Fallback.ServiceTier,
 			ProviderConfig:    m.Fallback.ProviderConfig.toDomain(),
 		}
 	}
@@ -418,6 +420,7 @@ type runtimePublicModel struct {
 	Description                          *string                `json:"description,omitempty"`
 	ProviderModelID                      string                 `json:"provider_model_id"`
 	UpstreamModelName                    string                 `json:"upstream_model_name"`
+	ServiceTier                          *string                `json:"service_tier,omitempty"`
 	ProviderConfig                       runtimeProviderConfig  `json:"provider_config"`
 	Fallback                             *runtimeProviderTarget `json:"fallback,omitempty"`
 	SupportsChatCompletions              *bool                  `json:"supports_chat_completions"`
@@ -440,6 +443,7 @@ type runtimePublicModel struct {
 type runtimeProviderTarget struct {
 	ProviderModelID   string                `json:"provider_model_id"`
 	UpstreamModelName string                `json:"upstream_model_name"`
+	ServiceTier       *string               `json:"service_tier,omitempty"`
 	ProviderConfig    runtimeProviderConfig `json:"provider_config"`
 }
 

apps/gateway/internal/adapters/metering/runtime_client_test.go

@@ -249,8 +249,10 @@ func TestRuntimeClientRejectsInvalidModelSuccess(t *testing.T) {
 
 func TestRuntimeClientMapsProviderFallback(t *testing.T) {
 	response := validRuntimePublicModel("openai/gpt-test")
+	response.ServiceTier = stringPointer("priority")
 	response.Fallback = &runtimeProviderTarget{
 		ProviderModelID: "fallback-model", UpstreamModelName: "fallback-upstream",
+		ServiceTier: stringPointer("flex"),
 		ProviderConfig: runtimeProviderConfig{
 			ID: "fallback-config", ProviderName: "fallback", BaseURL: "https://fallback.test",
 			APIKeySecretRef: "FALLBACK_API_KEY", Active: true,
@@ -267,6 +269,10 @@ func TestRuntimeClientMapsProviderFallback(t *testing.T) {
 		model.Fallback.ProviderConfig.ProviderName != "fallback" {
 		t.Fatalf("fallback = %#v", model.Fallback)
 	}
+	if model.ServiceTier == nil || *model.ServiceTier != "priority" ||
+		model.Fallback.ServiceTier == nil || *model.Fallback.ServiceTier != "flex" {
+		t.Fatalf("service tiers = primary %v, fallback %v", model.ServiceTier, model.Fallback.ServiceTier)
+	}
 }
 
 func TestRuntimeClientModelListRequiresArrayAndValidUniqueItems(t *testing.T) {
@@ -566,6 +572,10 @@ func boolPointer(value bool) *bool {
 	return &value
 }
 
+func stringPointer(value string) *string {
+	return &value
+}
+
 func int64Pointer(value int64) *int64 {
 	return &value
 }

apps/gateway/internal/adapters/providers/openai/proxy.go

@@ -6,6 +6,7 @@ import (
 	"encoding/json"
 	"io"
 	"net/http"
+	"strings"
 
 	"github.com/dappnode/dappnode-nexus-gateway/apps/gateway/internal/adapters/observability/metrics"
 	"github.com/dappnode/dappnode-nexus-gateway/apps/gateway/internal/application/ports"
@@ -22,6 +23,7 @@ var gatewayOnlyFields = []string{"provider_options"}
 // provider extensions). The only edits are:
 //
 //   - model: the provider's name for the public model;
+//   - service_tier: fixed by the catalog for Doubleword models;
 //   - stream_options.include_usage: on for streams, so usage can be metered;
 //   - the output-token limit: clamped to the model's maximum, and sent as
 //     max_tokens to the providers that only document that name (Novita,
@@ -41,6 +43,13 @@ func PrepareProxyBody(raw []byte, model domain.PublicModel, stream bool) (map[st
 	var transforms []string
 
 	body["model"] = model.UpstreamModelName
+	// Doubleword's realtime and async catalog entries select the upstream tier.
+	// Never let a caller change the tier behind a priced public model.
+	if model.ServiceTier != nil && strings.TrimSpace(*model.ServiceTier) != "" {
+		body["service_tier"] = strings.TrimSpace(*model.ServiceTier)
+	} else if strings.EqualFold(model.ProviderConfig.ProviderName, "doubleword") {
+		delete(body, "service_tier")
+	}
 	for _, field := range gatewayOnlyFields {
 		delete(body, field)
 	}

apps/gateway/internal/adapters/providers/openai/proxy_test.go

@@ -56,6 +56,41 @@ func TestPrepareProxyBody_ForwardsTheClientRequest(t *testing.T) {
 	}
 }
 
+func TestPrepareProxyBody_DoublewordServiceTier(t *testing.T) {
+	for _, stream := range []bool{false, true} {
+		for _, tc := range []struct {
+			name       string
+			provider   string
+			configured *string
+			client     string
+			want       string
+			present    bool
+		}{
+			{name: "async", provider: "doubleword", configured: stringPtr(" flex "), client: "priority", want: "flex", present: true},
+			{name: "realtime", provider: "doubleword", client: "priority"},
+			{name: "other provider", provider: "openai", client: "priority", want: "priority", present: true},
+		} {
+			t.Run(tc.name, func(t *testing.T) {
+				raw := `{"model":"public/model","messages":[],"service_tier":"` + tc.client + `"}`
+				body, _, err := PrepareProxyBody([]byte(raw), domain.PublicModel{
+					UpstreamModelName: "upstream",
+					ProviderConfig:    domain.ProviderConfig{ProviderName: tc.provider},
+					ServiceTier:       tc.configured,
+				}, stream)
+				if err != nil {
+					t.Fatal(err)
+				}
+				got, ok := body["service_tier"]
+				if ok != tc.present || ok && got != tc.want {
+					t.Fatalf("service_tier = %v (present %v), want %q (present %v)", got, ok, tc.want, tc.present)
+				}
+			})
+		}
+	}
+}
+
+func stringPtr(value string) *string { return &value }
+
 func TestPrepareProxyBody_Edits(t *testing.T) {
 	t.Run("streams always report usage", func(t *testing.T) {
 		got := prepared(t, `{"messages":[],"stream":true,"stream_options":{"include_usage":false,"continuous_usage_stats":true}}`, "novita", true)

apps/gateway/internal/application/services/generate_service.go

@@ -81,6 +81,7 @@ func shouldTryFallback(ctx context.Context, err error, fallback *domain.Provider
 func withProviderTarget(model domain.PublicModel, target domain.ProviderTarget) domain.PublicModel {
 	model.ProviderModelID = target.ProviderModelID
 	model.UpstreamModelName = target.UpstreamModelName
+	model.ServiceTier = target.ServiceTier
 	model.ProviderConfig = target.ProviderConfig
 	model.Fallback = nil
 	return model
@@ -202,6 +203,10 @@ func (s *GenerateService) validateRequest(endpoint string, req domain.GenerateRe
 		return domain.ErrUnsupportedFeature("structured_output")
 	}
 
+	if strings.EqualFold(model.ProviderConfig.ProviderName, "doubleword") && req.ServiceTier != nil {
+		return domain.ErrInvalidField("service_tier is fixed by the selected Doubleword model")
+	}
+
 	if model.EffectiveProofMode() == domain.ProofModeTinfoilAttestedTransport &&
 		!strings.EqualFold(model.ProviderConfig.ProviderName, "tinfoil") {
 		return domain.ErrUnsupportedFeature("Tinfoil verified transport for non-Tinfoil provider")

apps/gateway/internal/application/services/generate_service_test.go

@@ -239,6 +239,9 @@ func TestProxy_UsesConfiguredFallbackOnce(t *testing.T) {
 	if meter.lastSuccessModel.ProviderConfig.ProviderName != "fallback" {
 		t.Fatalf("metered provider = %q, want fallback", meter.lastSuccessModel.ProviderConfig.ProviderName)
 	}
+	if meter.lastSuccessModel.ServiceTier == nil || *meter.lastSuccessModel.ServiceTier != "flex" {
+		t.Fatalf("metered service tier = %v, want flex", meter.lastSuccessModel.ServiceTier)
+	}
 }
 
 func TestProxy_ReturnsFallbackFailure(t *testing.T) {
@@ -299,6 +302,7 @@ func newDirectModelGenerateService(meter *stubUsageMeter, provider *stubProvider
 }
 
 func newFallbackGenerateService(meter *stubUsageMeter, primary, fallback *stubProvider) *GenerateService {
+	serviceTier := "flex"
 	return NewGenerateService(
 		&stubAuthService{authCtx: domain.AuthContext{
 			Account: domain.Account{ID: "acc1", Status: domain.AccountStatusActive},
@@ -309,7 +313,7 @@ func newFallbackGenerateService(meter *stubUsageMeter, primary, fallback *stubPr
 			ProviderModelID:               "primary-model",
 			UpstreamModelName:             "primary-upstream",
 			ProviderConfig:                domain.ProviderConfig{ProviderName: "primary"},
-			Fallback:                      &domain.ProviderTarget{ProviderModelID: "fallback-model", UpstreamModelName: "fallback-upstream", ProviderConfig: domain.ProviderConfig{ProviderName: "fallback"}},
+			Fallback:                      &domain.ProviderTarget{ProviderModelID: "fallback-model", UpstreamModelName: "fallback-upstream", ServiceTier: &serviceTier, ProviderConfig: domain.ProviderConfig{ProviderName: "fallback"}},
 			SupportsChatCompletions:       true,
 			SupportsChatCompletionsStream: true,
 			MaxContextWindow:              1000,
@@ -377,6 +381,33 @@ func TestProxy_RejectsWhenBalanceIsEmpty(t *testing.T) {
 	}
 }
 
+func TestValidateRequest_RejectsCallerSelectedDoublewordServiceTier(t *testing.T) {
+	svc := &GenerateService{}
+	serviceTier := "flex"
+	model := domain.PublicModel{
+		PublicModelID:           "doubleword/test-realtime",
+		SupportsChatCompletions: true,
+		ProviderConfig:          domain.ProviderConfig{ProviderName: "doubleword"},
+	}
+	err := svc.validateRequest(domain.EndpointChatCompletions, domain.GenerateRequest{
+		ServiceTier: &serviceTier,
+	}, model)
+	if err == nil {
+		t.Fatal("expected caller-selected Doubleword service tier to be rejected")
+	}
+	gwErr, ok := err.(*domain.GatewayError)
+	if !ok || gwErr.Code != domain.ErrCodeInvalidField {
+		t.Fatalf("error = %#v, want invalid_field", err)
+	}
+
+	model.ProviderConfig.ProviderName = "openai"
+	if err := svc.validateRequest(domain.EndpointChatCompletions, domain.GenerateRequest{
+		ServiceTier: &serviceTier,
+	}, model); err != nil {
+		t.Fatalf("non-Doubleword service tier rejected: %v", err)
+	}
+}
+
 func TestProxy_DoesNotRouteUnknownModel(t *testing.T) {
 	router := &stubRouterClient{decision: domain.RouteDecision{PublicModelID: "minimax/minimax-m2.7"}}
 	svc := NewGenerateService(

pkg/domain/generation.go

@@ -28,6 +28,7 @@ type GenerateRequest struct {
 
 	Stream            bool
 	MaxOutputTokens   *int
+	ServiceTier       *string
 	ParallelToolCalls *bool
 	// StructuredOutput is set when the client asks for JSON output.
 	StructuredOutput bool

pkg/domain/model.go

@@ -21,6 +21,7 @@ type ProviderConfig struct {
 type ProviderTarget struct {
 	ProviderModelID   string
 	UpstreamModelName string
+	ServiceTier       *string
 	ProviderConfig    ProviderConfig
 }
 
@@ -32,6 +33,7 @@ type PublicModel struct {
 	Description                   *string
 	ProviderModelID               string
 	UpstreamModelName             string
+	ServiceTier                   *string
 	ProviderConfig                ProviderConfig
 	Fallback                      *ProviderTarget
 	SupportsChatCompletions       bool