feat: capabilities in /v1/models, replacing features

README.md

@@ -73,7 +73,10 @@ print(response.choices[0].message.content)
 Supported capabilities depend on the selected model and can include streaming,
 tool calls, parallel tool calls, structured outputs and reasoning. The current
 capabilities for each model are shown in the
-[model catalog](https://nexus.dappnode.com/models). They describe each model
+[model catalog](https://nexus.dappnode.com/models). `GET /v1/models` lists
+them per model in `capabilities`, using models.dev field names (`modalities`,
+`tool_call`, `structured_output`, `reasoning`, `reasoning_options`, …); a
+missing field means not stated. They describe each model
 and are not checked against chat requests: the selected provider validates chat
 options, so a stale catalog flag cannot reject a request the provider would
 accept. Endpoint routing, privacy policy and credit checks still apply.

apps/gateway/internal/adapters/http/dto/models_response.go

@@ -1,5 +1,7 @@
 package dto
 
+import "encoding/json"
+
 // ModelsResponse is the response envelope for GET /v1/models.
 type ModelsResponse struct {
 	Object string      `json:"object"`
@@ -27,8 +29,10 @@ type ModelData struct {
 	OutputPricePer1MTokensUSD       *float64 `json:"output_price_per_1m_tokens_usd,omitempty"`
 	CacheReadPricePer1MTokensUSD    *float64 `json:"cache_read_price_per_1m_tokens_usd,omitempty"`
 	CacheWritePricePer1MTokensUSD   *float64 `json:"cache_write_price_per_1m_tokens_usd,omitempty"`
-	Features                        []string `json:"features"`
-	Endpoints                       []string `json:"endpoints"`
-	ProofMode                       string   `json:"proof_mode"`
-	ProofsEnabled                   bool     `json:"proofs_enabled"`
+	// Capabilities is what the operator states the model accepts, using
+	// models.dev field names. A missing field means not stated.
+	Capabilities  json.RawMessage `json:"capabilities,omitempty"`
+	Endpoints     []string        `json:"endpoints"`
+	ProofMode     string          `json:"proof_mode"`
+	ProofsEnabled bool            `json:"proofs_enabled"`
 }

apps/gateway/internal/adapters/http/handlers/handlers_test.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"net/http"
 	"net/http/httptest"
+	"reflect"
 	"strings"
 	"testing"
 
@@ -110,20 +111,18 @@ func TestModelsHandler_ListsModels(t *testing.T) {
 	catalog := &mockModelCatalog{
 		models: []domain.PublicModel{
 			{
-				PublicModelID:                 "openai/gpt-4.1-mini",
-				DisplayName:                   "GPT-4.1 Mini",
-				Description:                   &desc,
-				ProviderConfig:                domain.ProviderConfig{ProviderName: "openai"},
-				Active:                        true,
-				SupportsChatCompletions:       true,
-				SupportsChatCompletionsStream: true,
-				SupportsTools:                 true,
-				SupportsStructuredOutput:      true,
-				MaxContextWindow:              128000,
-				MaxOutputTokens:               16000,
-				Currency:                      "EUR",
-				InputPricePerMillion:          decimal.NewFromFloat(0.4),
-				OutputPricePerMillion:         decimal.NewFromFloat(1.6),
+				PublicModelID:           "openai/gpt-4.1-mini",
+				DisplayName:             "GPT-4.1 Mini",
+				Description:             &desc,
+				ProviderConfig:          domain.ProviderConfig{ProviderName: "openai"},
+				Active:                  true,
+				SupportsChatCompletions: true,
+				Capabilities:            json.RawMessage(`{"streaming":true,"tool_call":true,"structured_output":true}`),
+				MaxContextWindow:        128000,
+				MaxOutputTokens:         16000,
+				Currency:                "EUR",
+				InputPricePerMillion:    decimal.NewFromFloat(0.4),
+				OutputPricePerMillion:   decimal.NewFromFloat(1.6),
 			},
 			{
 				PublicModelID:           "anthropic/claude-sonnet-4",
@@ -154,20 +153,20 @@ func TestModelsHandler_ListsModels(t *testing.T) {
 	var body struct {
 		Object string `json:"object"`
 		Data   []struct {
-			ID                          string   `json:"id"`
-			Object                      string   `json:"object"`
-			OwnedBy                     string   `json:"owned_by"`
-			DisplayName                 string   `json:"display_name"`
-			Description                 *string  `json:"description,omitempty"`
-			ContextSize                 int      `json:"context_size"`
-			MaxOutputTokens             int      `json:"max_output_tokens"`
-			Currency                    string   `json:"currency"`
-			InputPricePer1MTokensCents  int64    `json:"input_price_per_1m_tokens_cents"`
-			OutputPricePer1MTokensCents int64    `json:"output_price_per_1m_tokens_cents"`
-			InputPricePer1MTokensUSD    float64  `json:"input_price_per_1m_tokens_usd"`
-			OutputPricePer1MTokensUSD   float64  `json:"output_price_per_1m_tokens_usd"`
-			Features                    []string `json:"features"`
-			Endpoints                   []string `json:"endpoints"`
+			ID                          string          `json:"id"`
+			Object                      string          `json:"object"`
+			OwnedBy                     string          `json:"owned_by"`
+			DisplayName                 string          `json:"display_name"`
+			Description                 *string         `json:"description,omitempty"`
+			ContextSize                 int             `json:"context_size"`
+			MaxOutputTokens             int             `json:"max_output_tokens"`
+			Currency                    string          `json:"currency"`
+			InputPricePer1MTokensCents  int64           `json:"input_price_per_1m_tokens_cents"`
+			OutputPricePer1MTokensCents int64           `json:"output_price_per_1m_tokens_cents"`
+			InputPricePer1MTokensUSD    float64         `json:"input_price_per_1m_tokens_usd"`
+			OutputPricePer1MTokensUSD   float64         `json:"output_price_per_1m_tokens_usd"`
+			Capabilities                map[string]bool `json:"capabilities"`
+			Endpoints                   []string        `json:"endpoints"`
 		} `json:"data"`
 	}
 	json.NewDecoder(resp.Body).Decode(&body)
@@ -212,18 +211,8 @@ func TestModelsHandler_ListsModels(t *testing.T) {
 	if first.OutputPricePer1MTokensUSD != 1.728 {
 		t.Errorf("data[0].output_price_usd = %v, want 1.728", first.OutputPricePer1MTokensUSD)
 	}
-	wantFeatures := map[string]bool{
-		"streaming":          true,
-		"function-calling":   true,
-		"structured-outputs": true,
-	}
-	if len(first.Features) != len(wantFeatures) {
-		t.Errorf("data[0].features = %v, want %v", first.Features, wantFeatures)
-	}
-	for _, f := range first.Features {
-		if !wantFeatures[f] {
-			t.Errorf("unexpected feature %q in data[0].features", f)
-		}
+	if want := map[string]bool{"streaming": true, "tool_call": true, "structured_output": true}; !reflect.DeepEqual(first.Capabilities, want) {
+		t.Errorf("data[0].capabilities = %v, want %v", first.Capabilities, want)
 	}
 	if len(first.Endpoints) != 1 || first.Endpoints[0] != "chat/completions" {
 		t.Errorf("data[0].endpoints = %v, want [chat/completions]", first.Endpoints)

apps/gateway/internal/adapters/http/handlers/models_handler.go

@@ -1,6 +1,7 @@
 package handlers
 
 import (
+	"encoding/json"
 	"net/http"
 	"strings"
 
@@ -72,7 +73,7 @@ func toPublicModelData(m domain.PublicModel, eurToUSDRate float64) dto.ModelData
 		OutputPricePer1MTokensUSD:       priceToUSDPtr(m.Currency, m.OutputPricePerMillion, eurToUSDRate),
 		CacheReadPricePer1MTokensUSD:    pricePtrToUSDPtr(m.Currency, m.CacheReadPricePerMillion, eurToUSDRate),
 		CacheWritePricePer1MTokensUSD:   pricePtrToUSDPtr(m.Currency, m.CacheWritePricePerMillion, eurToUSDRate),
-		Features:                        modelFeatures(m),
+		Capabilities:                    publishedCapabilities(m),
 		Endpoints:                       modelEndpoints(m),
 		ProofMode:                       proofMode,
 		ProofsEnabled:                   domain.ProofModeEnabled(proofMode),
@@ -88,37 +89,19 @@ func toRouterData(r domain.RouterEntry) dto.ModelData {
 		Created:     r.CreatedAt.Unix(),
 		DisplayName: r.DisplayName,
 		Description: r.Description,
-		Features:    []string{"routing"},
 		Endpoints:   []string{"chat/completions"},
 		ProofMode:   domain.ProofModeNone,
 	}
 }
 
-func modelFeatures(m domain.PublicModel) []string {
-	if m.SupportsEndpoint(domain.EndpointSystemOne) {
-		return []string{"structured-outputs"}
-	}
-	features := make([]string, 0, 5)
-	if m.SupportsChatCompletionsStream {
-		features = append(features, "streaming")
-	}
-	if m.SupportsTools {
-		features = append(features, "function-calling")
-	}
-	if m.SupportsParallelToolCalls {
-		features = append(features, "parallel-tool-calls")
-	}
-	if m.SupportsStructuredOutput {
-		features = append(features, "structured-outputs")
-	}
-	if m.SupportsReasoning {
-		features = append(features, "reasoning")
-	}
-	switch m.EffectiveProofMode() {
-	case domain.ProofModeTinfoilAttestedTransport:
-		features = append(features, "tinfoil-attested-transport")
+// publishedCapabilities is the operator's record, unchanged, or nothing when
+// the operator stated nothing.
+func publishedCapabilities(m domain.PublicModel) json.RawMessage {
+	var object map[string]json.RawMessage
+	if json.Unmarshal(m.Capabilities, &object) != nil || len(object) == 0 {
+		return nil
 	}
-	return features
+	return m.Capabilities
 }
 
 func modelEndpoints(m domain.PublicModel) []string {

apps/gateway/internal/adapters/http/handlers/models_handler_test.go

@@ -0,0 +1,28 @@
+package handlers
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+
+	"github.com/dappnode/dappnode-nexus-gateway/pkg/domain"
+)
+
+// The operator's record is published unchanged, including fields this Gateway
+// does not know yet. There is no separate features list any more.
+func TestModelsHandler_PublishesCapabilitiesUnchanged(t *testing.T) {
+	stored := `{"modalities":{"input":["text","image"],"output":["text"]},"tool_call":true,"reasoning":true,` +
+		`"reasoning_options":[{"type":"effort","values":["low","high","max"],"default":"high"}],"audio_output":false}`
+	data := toPublicModelData(domain.PublicModel{PublicModelID: "vision", SupportsChatCompletions: true, Capabilities: json.RawMessage(stored)}, 1)
+	if string(data.Capabilities) != stored {
+		t.Fatalf("capabilities = %s", data.Capabilities)
+	}
+	if body, _ := json.Marshal(data); strings.Contains(string(body), `"features"`) {
+		t.Fatalf("published the removed features list: %s", body)
+	}
+	empty := toPublicModelData(domain.PublicModel{PublicModelID: "unknown", SupportsChatCompletions: true, Capabilities: json.RawMessage(`{}`)}, 1)
+	body, _ := json.Marshal(empty)
+	if strings.Contains(string(body), `"capabilities"`) {
+		t.Fatalf("published capabilities nobody stated: %s", body)
+	}
+}

apps/gateway/internal/adapters/http/handlers/stream_usage_repro_test.go

@@ -107,7 +107,7 @@ func TestStreamingUsageRetention(t *testing.T) {
 
 			model := domain.PublicModel{
 				PublicModelID: "local-repro", ProviderModelID: "local-repro", UpstreamModelName: "local-repro", Active: true,
-				SupportsChatCompletions: true, SupportsChatCompletionsStream: true, SupportsTools: true,
+				SupportsChatCompletions: true, Capabilities: json.RawMessage(`{"streaming":true,"tool_call":true}`),
 				ProviderConfig: domain.ProviderConfig{
 					ProviderName: "novita", BaseURL: upstream.URL,
 					APIKeySecretRef: "NEXUS_STREAM_REPRO_PROVIDER_KEY",

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

@@ -322,29 +322,41 @@ func jsonObject(raw json.RawMessage) bool {
 	return json.Unmarshal(raw, &object) == nil && object != nil
 }
 
+// capabilities is the operator's record from Metering. A Metering that does not
+// send one yet still sends the flags it replaced, so they are used instead.
+func (m runtimePublicModel) capabilities() json.RawMessage {
+	if jsonObject(m.Capabilities) {
+		return m.Capabilities
+	}
+	legacy, _ := json.Marshal(map[string]bool{
+		"streaming":           *m.SupportsChatCompletionsStream,
+		"tool_call":           *m.SupportsTools,
+		"parallel_tool_calls": *m.SupportsParallelToolCalls,
+		"structured_output":   *m.SupportsStructuredOutput,
+		"reasoning":           *m.SupportsReasoning,
+	})
+	return legacy
+}
+
 func (m runtimePublicModel) toDomain() domain.PublicModel {
 	model := domain.PublicModel{
-		ID:                            m.ID,
-		PublicModelID:                 m.PublicModelID,
-		DisplayName:                   m.DisplayName,
-		Description:                   m.Description,
-		ProviderModelID:               m.ProviderModelID,
-		UpstreamModelName:             m.UpstreamModelName,
-		ServiceTier:                   m.ServiceTier,
-		ProviderConfig:                m.ProviderConfig.toDomain(),
-		SupportsChatCompletions:       *m.SupportsChatCompletions,
-		SupportsChatCompletionsStream: *m.SupportsChatCompletionsStream,
-		SupportsTools:                 *m.SupportsTools,
-		SupportsParallelToolCalls:     *m.SupportsParallelToolCalls,
-		SupportsStructuredOutput:      *m.SupportsStructuredOutput,
-		SupportsReasoning:             *m.SupportsReasoning,
-		ProofMode:                     m.ProofMode,
-		MaxContextWindow:              m.MaxContextWindow,
-		MaxOutputTokens:               m.MaxOutputTokens,
-		InputPricePerMillion:          priceFromMicrocents(*m.InputPricePer1MTokensMicrocents),
-		OutputPricePerMillion:         priceFromMicrocents(*m.OutputPricePer1MTokensMicrocents),
-		Currency:                      m.Currency,
-		Active:                        m.Active,
+		ID:                      m.ID,
+		PublicModelID:           m.PublicModelID,
+		DisplayName:             m.DisplayName,
+		Description:             m.Description,
+		ProviderModelID:         m.ProviderModelID,
+		UpstreamModelName:       m.UpstreamModelName,
+		ServiceTier:             m.ServiceTier,
+		ProviderConfig:          m.ProviderConfig.toDomain(),
+		SupportsChatCompletions: *m.SupportsChatCompletions,
+		Capabilities:            m.capabilities(),
+		ProofMode:               m.ProofMode,
+		MaxContextWindow:        m.MaxContextWindow,
+		MaxOutputTokens:         m.MaxOutputTokens,
+		InputPricePerMillion:    priceFromMicrocents(*m.InputPricePer1MTokensMicrocents),
+		OutputPricePerMillion:   priceFromMicrocents(*m.OutputPricePer1MTokensMicrocents),
+		Currency:                m.Currency,
+		Active:                  m.Active,
 	}
 	if m.Fallback != nil {
 		model.Fallback = &domain.ProviderTarget{
@@ -429,6 +441,7 @@ type runtimePublicModel struct {
 	SupportsParallelToolCalls            *bool                  `json:"supports_parallel_tool_calls"`
 	SupportsStructuredOutput             *bool                  `json:"supports_structured_output"`
 	SupportsReasoning                    *bool                  `json:"supports_reasoning"`
+	Capabilities                         json.RawMessage        `json:"capabilities,omitempty"`
 	ProofMode                            string                 `json:"proof_mode"`
 	MaxContextWindow                     int                    `json:"max_context_window"`
 	MaxOutputTokens                      int                    `json:"max_output_tokens"`

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

@@ -579,3 +579,24 @@ func stringPointer(value string) *string {
 func int64Pointer(value int64) *int64 {
 	return &value
 }
+
+// The operator's record is used as sent; a Metering that does not send one yet
+// still sends the flags it replaced.
+func TestRuntimePublicModelCapabilities(t *testing.T) {
+	model := validRuntimePublicModel("openai/gpt-test")
+	model.Capabilities = json.RawMessage(`{"modalities":{"input":["text","image"]},"tool_call":true}`)
+	if got := string(model.toDomain().Capabilities); got != `{"modalities":{"input":["text","image"]},"tool_call":true}` {
+		t.Fatalf("capabilities = %s", got)
+	}
+	model.Capabilities = nil
+	var legacy map[string]bool
+	if err := json.Unmarshal(model.toDomain().Capabilities, &legacy); err != nil {
+		t.Fatal(err)
+	}
+	want := map[string]bool{"streaming": true, "tool_call": false, "parallel_tool_calls": false, "structured_output": false, "reasoning": false}
+	for key, value := range want {
+		if got, ok := legacy[key]; !ok || got != value {
+			t.Fatalf("capabilities from flags = %v", legacy)
+		}
+	}
+}

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

@@ -2,6 +2,7 @@ package services
 
 import (
 	"context"
+	"encoding/json"
 	"io"
 	"strings"
 	"testing"
@@ -309,15 +310,15 @@ func newFallbackGenerateService(meter *stubUsageMeter, primary, fallback *stubPr
 			APIKey:  domain.APIKey{ID: "key1", Active: true},
 		}},
 		&stubModelCatalog{model: domain.PublicModel{
-			PublicModelID:                 "model",
-			ProviderModelID:               "primary-model",
-			UpstreamModelName:             "primary-upstream",
-			ProviderConfig:                domain.ProviderConfig{ProviderName: "primary"},
-			Fallback:                      &domain.ProviderTarget{ProviderModelID: "fallback-model", UpstreamModelName: "fallback-upstream", ServiceTier: &serviceTier, ProviderConfig: domain.ProviderConfig{ProviderName: "fallback"}},
-			SupportsChatCompletions:       true,
-			SupportsChatCompletionsStream: true,
-			MaxContextWindow:              1000,
-			MaxOutputTokens:               100,
+			PublicModelID:           "model",
+			ProviderModelID:         "primary-model",
+			UpstreamModelName:       "primary-upstream",
+			ProviderConfig:          domain.ProviderConfig{ProviderName: "primary"},
+			Fallback:                &domain.ProviderTarget{ProviderModelID: "fallback-model", UpstreamModelName: "fallback-upstream", ServiceTier: &serviceTier, ProviderConfig: domain.ProviderConfig{ProviderName: "fallback"}},
+			SupportsChatCompletions: true,
+			Capabilities:            json.RawMessage(`{"streaming":true}`),
+			MaxContextWindow:        1000,
+			MaxOutputTokens:         100,
 		}},
 		nil,
 		&stubProviderRegistry{providers: map[string]ports.Provider{
@@ -456,16 +457,16 @@ func TestProxy_RoutesExplicitRouterToSelectedModel(t *testing.T) {
 		Reason:         reason,
 	}}
 	selected := domain.PublicModel{
-		PublicModelID:                 "minimax/minimax-m2.7",
-		DisplayName:                   "MiniMax M2.7",
-		ProviderModelID:               "novita-minimax-m2.7",
-		ProviderConfig:                domain.ProviderConfig{ProviderName: "novita"},
-		SupportsChatCompletions:       true,
-		SupportsChatCompletionsStream: true,
-		MaxContextWindow:              1000,
-		MaxOutputTokens:               100,
-		InputPricePerMillion:          decimal.NewFromFloat(0.03),
-		OutputPricePerMillion:         decimal.NewFromFloat(0.12),
+		PublicModelID:           "minimax/minimax-m2.7",
+		DisplayName:             "MiniMax M2.7",
+		ProviderModelID:         "novita-minimax-m2.7",
+		ProviderConfig:          domain.ProviderConfig{ProviderName: "novita"},
+		SupportsChatCompletions: true,
+		Capabilities:            json.RawMessage(`{"streaming":true}`),
+		MaxContextWindow:        1000,
+		MaxOutputTokens:         100,
+		InputPricePerMillion:    decimal.NewFromFloat(0.03),
+		OutputPricePerMillion:   decimal.NewFromFloat(0.12),
 	}
 	svc := NewGenerateService(
 		&stubAuthService{authCtx: domain.AuthContext{