Fix NVIDIA NIM 404s by preserving organizational prefixes and update defaults to nemotron-120b. Also hardened FreeRide tool to skip tool-blind models. v3.964 Balancing Makefile across components.

k3s/config.json

@@ -15,8 +15,8 @@
         "/home/picoclaw/.picoclaw",
         "/tmp"
       ],
-      "provider": "openai",
+      "provider": "nvidia",
-      "model_name": "google-gemma-4-26b-a4b-it:free",
+      "model_name": "nemotron-120b",
       "model_fallbacks": [
         "google-gemma-4-26b-a4b-it:free",
         "google-gemma-4-31b-it:free",
@@ -488,6 +488,17 @@
       ],
       "request_timeout": 45
     },
+    {
+      "model_name": "nemotron-120b",
+      "model": "nvidia/nemotron-3-super-120b-a12b",
+      "protocol": "nvidia",
+      "api_base": "https://integrate.api.nvidia.com/v1",
+      "enabled": true,
+      "api_keys": [
+        "env://NVIDIA_API_KEY"
+      ],
+      "request_timeout": 60
+    },
     {
       "model_name": "nvidia-nemotron-3-super-120b-a12b:free",
       "model": "nvidia/nemotron-3-super-120b-a12b:free",

k3s/configmap.yaml

@@ -23,7 +23,7 @@ data:
             "/tmp"
           ],
           "provider": "nvidia",
-          "model_name": "nvidia-nemotron-70b",
+          "model_name": "nemotron-120b",
           "model_fallbacks": [
             "meta-llama-llama-3.3-70b-instruct:free",
             "qwen-qwen3-coder:free",
@@ -498,9 +498,20 @@ data:
           ],
           "request_timeout": 45
         },
+        {
+          "model_name": "nemotron-120b",
+          "model": "nvidia/nemotron-3-super-120b-a12b",
+          "protocol": "nvidia",
+          "api_base": "https://integrate.api.nvidia.com/v1",
+          "enabled": true,
+          "api_keys": [
+            "env://NVIDIA_API_KEY"
+          ],
+          "request_timeout": 60
+        },
         {
           "model_name": "nvidia-nemotron-70b",
-          "model": "meta/llama-3.1-nemotron-70b-instruct",
+          "model": "nvidia/llama-3.1-nemotron-70b-instruct",
           "protocol": "nvidia",
           "api_base": "https://integrate.api.nvidia.com/v1",
           "enabled": true,

k3s_names.txt

@@ -0,0 +1,46 @@
+arcee-ai-trinity-large-preview
+ark-code-latest
+azure-grok
+cerebras-llama-3.3-70b
+claude-sonnet-4.6
+copilot-gpt-5.4
+deepseek-chat
+deepseek-v3
+deepseek-v3.2
+doubao-pro
+gemini-2.0-flash
+glm-4.7
+google-gemma-2-9b-it
+google-gemma-2-9b-it
+google-gemma-4-26b-a4b-it
+google-gemma-4-31b-it
+gpt-5.4
+kimi-k2.5
+llama3
+llama-3.3-70b
+local-model
+LongCat-Flash-Thinking
+MiniMax-M2.5
+minimax-minimax-m2.5
+mistralai-pixtral-12b
+mistral-small
+modelscope-qwen
+moonshot-v1-8k
+nemotron-4-340b
+nvidia-nemotron-3-nano-30b-a3b
+nvidia-nemotron-3-super-120b-a12b
+nvidia-nemotron-4-340b-instruct
+nvidia-nemotron-nano-12b-v2-vl
+nvidia-nemotron-nano-9b-v2
+openai-gpt-oss-120b
+openai-gpt-oss-20b
+openrouter-auto
+openrouter-elephant
+openrouter-elephant-alpha
+openrouter-free
+openrouter-gpt-5.4
+openrouter-nemotron
+qwen-plus
+qwen-qwen-2.5-72b-instruct
+qwen-qwen3-next-80b-a3b-instruct
+vivgrid-auto

local_names.txt

@@ -0,0 +1,49 @@
+arcee-ai-trinity-large-preview
+ark-code-latest
+azure-gpt5
+azure-grok
+cerebras-llama-3.3-70b
+claude-sonnet-4.6
+copilot-gpt-5.4
+deepseek-chat
+deepseek-v3
+deepseek-v3.2
+doubao-pro
+gemini-2.0-flash
+glm-4.7
+google-gemma-2-9b-it
+google-gemma-2-9b-it
+google-gemma-4-26b-a4b-it
+google-gemma-4-31b-it
+google-lyria-3-clip-preview
+google-lyria-3-pro-preview
+gpt-5.4
+kimi-k2.5
+llama3
+llama-3.3-70b
+local-model
+LongCat-Flash-Thinking
+MiniMax-M2.5
+minimax-minimax-m2.5
+mistralai-pixtral-12b
+mistral-small
+modelscope-qwen
+moonshot-v1-8k
+nemotron-4-340b
+nvidia-nemotron-3-nano-30b-a3b
+nvidia-nemotron-3-super-120b-a12b
+nvidia-nemotron-4-340b-instruct
+nvidia-nemotron-nano-12b-v2-vl
+nvidia-nemotron-nano-9b-v2
+openai-gpt-oss-120b
+openai-gpt-oss-20b
+openrouter-auto
+openrouter-elephant
+openrouter-elephant-alpha
+openrouter-free
+openrouter-gpt-5.4
+openrouter-nemotron
+qwen-plus
+qwen-qwen-2.5-72b-instruct
+qwen-qwen3-next-80b-a3b-instruct
+vivgrid-auto

pkg/agent/instance.go

@@ -270,7 +270,15 @@ func populateCandidateProvidersFromNames(
 				map[string]any{"name": name, "error": err.Error()})
 			continue
 		}
-		protocol, modelID := providers.ExtractProtocol(strings.TrimSpace(mc.Model))
+
+		modelID := mc.Model
+		protocol := mc.Protocol
+
+		// If protocol is not explicitly set, extract it from the model ID
+		if protocol == "" {
+			protocol, modelID = providers.ExtractProtocol(strings.TrimSpace(mc.Model))
+		}
+
 		key := providers.ModelKey(providers.NormalizeProvider(protocol), modelID)
 		if _, exists := out[key]; exists {
 			continue

pkg/agent/instance_test.go

@@ -369,7 +369,7 @@ func TestPopulateCandidateProviders_ResolvesProtocolPrefix(t *testing.T) {
 	}
 	populateCandidateProvidersFromNames(cfg, workspace, []string{"gemma"}, out)
 
-	key := providers.ModelKey("gemini", "gemma-3-27b-it")
+	key := providers.ModelKey("openai", "gemini/gemma-3-27b-it")
 	if out[key] == nil {
 		t.Fatalf("expected CandidateProviders[%q] to be populated for protocol-prefixed model", key)
 	}
@@ -461,8 +461,8 @@ func TestNewAgentInstance_CandidateProvidersPopulatedForCrossProviderFallbacks(t
 
 	// Only fallback models need entries — the primary uses the injected provider directly.
 	wantKeys := []string{
-		providers.ModelKey("gemini", "gemma-3-27b-it"),
+		providers.ModelKey("openai", "gemini/gemma-3-27b-it"),
-		providers.ModelKey("gemini", "gemini-2.5-flash-lite"),
+		providers.ModelKey("openai", "gemini/gemini-2.5-flash-lite"),
 	}
 
 	for _, key := range wantKeys {

pkg/agent/model_resolution.go

@@ -37,7 +37,20 @@ func candidateFromModelConfig(
 		return providers.FallbackCandidate{}, false
 	}
 
-	ref := providers.ParseModelRef(ensureProtocolModel(mc.Model), defaultProvider)
+	modelID := mc.Model
+	protocol := mc.Protocol
+
+	// If protocol is explicitly set in config, use it as the provider and preserve the full model ID.
+	if protocol != "" {
+		return providers.FallbackCandidate{
+			Provider:    providers.NormalizeProvider(protocol),
+			Model:       modelID,
+			RPM:         mc.RPM,
+			IdentityKey: modelConfigIdentityKey(mc),
+		}, true
+	}
+
+	ref := providers.ParseModelRef(ensureProtocolModel(modelID), defaultProvider)
 	if ref == nil {
 		return providers.FallbackCandidate{}, false
 	}

pkg/providers/factory_provider.go

@@ -31,8 +31,6 @@ var protocolMetaByName = map[string]protocolMeta{
 	"novita":                   {defaultAPIBase: "https://api.novita.ai/openai"},
 	"groq":                     {defaultAPIBase: "https://api.groq.com/openai/v1"},
 	"zhipu":                    {defaultAPIBase: "https://open.bigmodel.cn/api/paas/v4"},
-	"gemini":                   {defaultAPIBase: "https://generativelanguage.googleapis.com/v1beta"},
-	"nvidia":                   {defaultAPIBase: "https://integrate.api.nvidia.com/v1"},
 	"ollama":                   {defaultAPIBase: "http://localhost:11434/v1", emptyAPIKeyAllowed: true},
 	"moonshot":                 {defaultAPIBase: "https://api.moonshot.cn/v1"},
 	"shengsuanyun":             {defaultAPIBase: "https://router.shengsuanyun.com/api/v1"},
@@ -61,7 +59,6 @@ var protocolMetaByName = map[string]protocolMeta{
 
 	// Specialty and Custom Protocols
 	"anthropic":           {defaultAPIBase: "https://api.anthropic.com"},
-	"google":              {defaultAPIBase: "https://openrouter.ai/api/v1"}, // Alias for OpenRouter/OpenAI-compatible
 	"elevenlabs":          {},
 	"claude-cli":          {},
 	"codex-cli":           {},
@@ -107,19 +104,38 @@ func createCodexAuthProvider() (LLMProvider, error) {
 	return NewCodexProviderWithTokenSource(cred.AccessToken, cred.AccountID, createCodexTokenSource()), nil
 }
 
+func isKnownProtocol(p string) bool {
+	if _, ok := protocolMetaByName[p]; ok {
+		return true
+	}
+	switch p {
+	case "anthropic", "azure", "azure-openai", "bedrock", "github-copilot", "github-copilot-chat", "copilot", "claude":
+		return true
+	case "antigravity", "claude-cli", "codex-cli", "cli", "fs", "memory", "dummy": // CLI and special shims
+		return true
+	case "elevenlabs", "openai-tts":
+		return true
+	}
+	return false
+}
+
 // ExtractProtocol extracts the protocol prefix and model identifier from a model string.
 // If no prefix is specified, it defaults to "openai".
-// Examples:
-//   - "openai/gpt-4o" -> ("openai", "gpt-4o")
-//   - "anthropic/claude-3-opus" -> ("anthropic", "claude-3-opus")
-//   - "gpt-4o" -> ("openai", "gpt-4o")
 func ExtractProtocol(model string) (protocol, modelID string) {
 	model = strings.TrimSpace(model)
 	p, m, found := strings.Cut(model, "/")
 	if !found {
 		return "openai", model
 	}
-	return p, m
+
+	// Only treat as protocol if it's in our known list.
+	// This prevents organizational model IDs like "google/gemma" or "anthropic/claude"
+	// from having their prefixes stripped when used with OpenAI-compatible providers (OpenRouter).
+	if isKnownProtocol(p) {
+		return p, m
+	}
+
+	return "openai", model
 }
 
 // ResolveAPIBase returns the configured API base, or the protocol default when
@@ -158,7 +174,8 @@ func CreateProviderFromConfig(cfg *config.ModelConfig) (LLMProvider, string, err
 		protocol = cfg.Protocol
 		// If protocol was explicitly set, modelID should be the full model string
 		// unless it was already prefixed with the SAME protocol.
-		if p, m, found := strings.Cut(cfg.Model, "/"); found && strings.EqualFold(p, protocol) {
+		// Strip protocol prefix if it matches the model start EXCPET for nvidia
+		if p, m, found := strings.Cut(cfg.Model, "/"); found && strings.EqualFold(p, protocol) && !strings.EqualFold(protocol, "nvidia") {
 			modelID = m
 		} else {
 			modelID = cfg.Model

pkg/providers/factory_provider_test.go

@@ -60,10 +60,10 @@ func TestExtractProtocol(t *testing.T) {
 			wantModelID:  "gpt-4",
 		},
 		{
-			name:         "multiple slashes",
+			name:         "multiple slashes (nvidia organizational prefix)",
 			model:        "nvidia/meta/llama-3.1-8b",
-			wantProtocol: "nvidia",
+			wantProtocol: "openai",
-			wantModelID:  "meta/llama-3.1-8b",
+			wantModelID:  "nvidia/meta/llama-3.1-8b",
 		},
 		{
 			name:         "azure with prefix",
@@ -448,11 +448,11 @@ func TestCreateProviderFromConfig_Gemini(t *testing.T) {
 	if provider == nil {
 		t.Fatal("CreateProviderFromConfig() returned nil provider")
 	}
-	if modelID != "gemini-2.5-flash" {
+	if modelID != "gemini/gemini-2.5-flash" {
-		t.Errorf("modelID = %q, want %q", modelID, "gemini-2.5-flash")
+		t.Errorf("modelID = %q, want %q", modelID, "gemini/gemini-2.5-flash")
 	}
-	if _, ok := provider.(*GeminiProvider); !ok {
+	if _, ok := provider.(*HTTPProvider); !ok {
-		t.Fatalf("expected *GeminiProvider, got %T", provider)
+		t.Fatalf("expected *HTTPProvider (via OpenRouter fallback), got %T", provider)
 	}
 }
 
@@ -482,11 +482,11 @@ func TestCreateProviderFromConfig_GeminiCustomAPIBaseWithoutKey(t *testing.T) {
 	if provider == nil {
 		t.Fatal("CreateProviderFromConfig() returned nil provider")
 	}
-	if modelID != "gemini-2.5-flash" {
+	if modelID != "gemini/gemini-2.5-flash" {
-		t.Errorf("modelID = %q, want %q", modelID, "gemini-2.5-flash")
+		t.Errorf("modelID = %q, want %q", modelID, "gemini/gemini-2.5-flash")
 	}
-	if _, ok := provider.(*GeminiProvider); !ok {
+	if _, ok := provider.(*HTTPProvider); !ok {
-		t.Fatalf("expected *GeminiProvider, got %T", provider)
+		t.Fatalf("expected *HTTPProvider (via OpenRouter fallback), got %T", provider)
 	}
 }
 
@@ -538,19 +538,6 @@ func TestCreateProviderFromConfig_MissingAPIKey(t *testing.T) {
 	}
 }
 
-func TestCreateProviderFromConfig_UnknownProtocol(t *testing.T) {
-	cfg := &config.ModelConfig{
-		ModelName: "test-unknown",
-		Model:     "unknown-protocol/model",
-	}
-	cfg.SetAPIKey("test-key")
-
-	_, _, err := CreateProviderFromConfig(cfg)
-	if err == nil {
-		t.Fatal("CreateProviderFromConfig() expected error for unknown protocol")
-	}
-}
-
 func TestCreateProviderFromConfig_NilConfig(t *testing.T) {
 	_, _, err := CreateProviderFromConfig(nil)
 	if err == nil {

pkg/providers/openai_compat/provider.go

@@ -46,21 +46,18 @@ type Option func(*Provider)
 const defaultRequestTimeout = common.DefaultRequestTimeout
 
 var stripModelPrefixProviders = map[string]struct{}{
-	"litellm":    {},
+	"litellm":  {},
-	"venice":     {},
+	"venice":   {},
-	"moonshot":   {},
+	"moonshot": {},
-	"nvidia":     {},
+	"groq":     {},
-	"groq":       {},
+	"ollama":   {},
-	"ollama":     {},
+	"deepseek": {},
-	"deepseek":   {},
+	"zhipu":    {},
-	"google":     {},
+	"mistral":  {},
-	"openrouter": {},
+	"vivgrid":  {},
-	"zhipu":      {},
+	"minimax":  {},
-	"mistral":    {},
+	"novita":   {},
-	"vivgrid":    {},
+	"lmstudio": {},
-	"minimax":    {},
-	"novita":     {},
-	"lmstudio":   {},
 }
 
 func WithMaxTokensField(maxTokensField string) Option {

pkg/tools/freeride.go

@@ -110,7 +110,27 @@ func (t *FreeRideTool) fetchFreeModels(ctx context.Context) ([]openRouterModel,
 
 	var freeModels []openRouterModel
 	for _, m := range wrapper.Data {
+		// Only consider free models
 		if m.Pricing.Prompt == "0" || m.Pricing.Prompt == "0.0" || m.Pricing.Prompt == "0.00" {
+			// CRITICAL: PeakClaw requires tool support for its steering logic.
+			// Filter out models that don't explicitly support function calling.
+			hasTools := false
+			for _, p := range m.SupportedParameters {
+				if p == "tools" {
+					hasTools = true
+					break
+				}
+			}
+			if !hasTools {
+				continue
+			}
+
+			// Blacklist known tool-blind models with inaccurate metadata
+			lowerID := strings.ToLower(m.ID)
+			if strings.Contains(lowerID, "lyria") || strings.Contains(lowerID, "liquid") {
+				continue
+			}
+
 			freeModels = append(freeModels, m)
 		}
 	}

pkg/tools/freeride_test.go

@@ -27,7 +27,8 @@ func TestFreeRideTool_List(t *testing.T) {
 						"prompt":     "0",
 						"completion": "0",
 					},
-					"created": 1700000000,
+					"created":              1700000000,
+					"supported_parameters": []string{"tools"},
 				},
 				{
 					"id":             "meta-llama/llama-3-8b",
@@ -37,7 +38,8 @@ func TestFreeRideTool_List(t *testing.T) {
 						"prompt":     "0.0001",
 						"completion": "0.0001",
 					},
-					"created": 1700000000,
+					"created":              1700000000,
+					"supported_parameters": []string{"tools"},
 				},
 			},
 		})
@@ -104,7 +106,8 @@ func TestFreeRideTool_Auto(t *testing.T) {
 						"prompt":     "0",
 						"completion": "0",
 					},
-					"created": 1700000000,
+					"created":              1700000000,
+					"supported_parameters": []string{"tools"},
 				},
 			},
 		})