@aria_asi/cli 0.2.36 → 0.2.38

package/runtime-src/local-phase.mjs

@@ -622,6 +622,7 @@ export function buildRuntimeCognitionDirective(packet, bundle = {}) {
     '- For any non-trivial reply, include a readable <cognition> block using everyday labels, not internal lens codenames.',
     '- If you request any non-trivial tool call, place the <cognition> block BEFORE the tool request.',
     '- If the action is deploy, destructive, or state-mutating, include both <verify> and <expected> blocks before the tool request.',
+    '- For any non-trivial reply, also include <applied_cognition> so the runtime can verify that cognition changed the response.',
     '- The runtime will canonicalize the readable block into backend JSON automatically, so do not hide it behind private shorthand.',
     '',
     '<cognition>',
@@ -635,6 +636,13 @@ export function buildRuntimeCognitionDirective(packet, bundle = {}) {
     '  beauty: <what cleaner, more elegant, more durable path preserves the contract>',
     `  first_principle: ${firstPrincipleText}`,
     '</cognition>',
+    '<applied_cognition>',
+    '  decision_delta: <specific change this cognition made to the answer>',
+    '  dominant_domain: <primary domain pressure shaping the response>',
+    '  binds_to: <one visible substrate, evidence, axiom, or user constraint>',
+    '  expected_predicate: <observable predicate that would prove the answer held up>',
+    '  artifact_change: <what changed in the emitted answer because of cognition>',
+    '</applied_cognition>',
     '',
     dominantModules.length ? 'Current module insights:' : 'Current module insights: none yet',
     ...dominantModules.map((line) => `- ${line}`),

package/runtime-src/provider-proxy.mjs

@@ -7,6 +7,7 @@ import { join } from 'node:path';
 const CONFIG_PATH = join(homedir(), '.aria', 'config.json');
 const DEFAULT_OLLAMA_BASE_URL = process.env.OLLAMA_BASE_URL || 'http://127.0.0.1:11434';
 const DEFAULT_DEEPSEEK_BASE_URL = process.env.DEEPSEEK_API_BASE || 'https://api.deepseek.com/v1/chat/completions';
+const DEFAULT_XAI_BASE_URL = process.env.XAI_API_BASE || 'https://api.x.ai/v1/chat/completions';
 const DEFAULT_OLLAMA_MODEL =
   process.env.ARIA_RUNTIME_MODEL ||
   process.env.ARIA_LOCAL_MODEL ||
@@ -15,18 +16,32 @@ const DEFAULT_OLLAMA_MODEL =
 const DEFAULT_CHAT_MODEL =
   process.env.ARIA_CHAT_MODEL ||
   process.env.ARIA_CASUAL_MODEL ||
-  'deepseek-v4-flash';
+  process.env.ARIA_XAI_MODEL ||
+  process.env.GROK_MODEL ||
+  'grok-4-3';
 const DEFAULT_DEEP_MODEL =
   process.env.ARIA_DEEP_MODEL ||
+  process.env.ARIA_DEEPSEEK_FALLBACK_MODEL ||
   'deepseek-v4-pro';
-const DEFAULT_XAI_BASE_URL = process.env.XAI_API_BASE || 'https://api.x.ai/v1/chat/completions';
 const DEFAULT_XAI_FALLBACK_MODEL =
   process.env.ARIA_XAI_FALLBACK_MODEL ||
-  'grok-4-2-reasoning';
+  DEFAULT_CHAT_MODEL;
 const DEFAULT_NIM_BASE_URL = process.env.NIM_BASE_URL || 'https://integrate.api.nvidia.com/v1/chat/completions';
 const DEFAULT_NIM_FALLBACK_MODEL =
   process.env.ARIA_NIM_FALLBACK_MODEL ||
   'qwen/qwen3.5-122b-a10b';
+const LOCAL_VIRTUAL_MAC_PROVIDER = 'aria-local-virtual-mac-lane';
+const DEFAULT_LOCAL_FALLBACK_BASE_URL = normalizeChatCompletionsUrl(
+  process.env.ARIA_LOCAL_FALLBACK_BASE_URL ||
+  process.env.ARIA_VIRTUAL_MAC_LANE_BASE_URL ||
+  process.env.ARIA_LANE_GATEWAY_URL ||
+  'http://aria-lane-gateway.aria.svc.cluster.local:8089/v1',
+);
+const DEFAULT_LOCAL_FALLBACK_MODEL =
+  process.env.ARIA_LOCAL_FALLBACK_MODEL ||
+  process.env.ARIA_VIRTUAL_MAC_LANE_MODEL ||
+  process.env.ARIA_LOCAL_MAC_MODEL ||
+  'mlx-community/Qwen2.5-7B-Instruct-4bit';
 
 function safeJsonParse(raw, fallback = null) {
   try {
@@ -45,6 +60,36 @@ function coerceString(value) {
   return typeof value === 'string' && value.trim() ? value.trim() : '';
 }
 
+function secretEnv(...names) {
+  for (const name of names) {
+    const value = coerceString(process.env[name]);
+    if (value) return value;
+  }
+  return '';
+}
+
+function normalizeChatCompletionsUrl(value) {
+  const trimmed = coerceString(value).replace(/\/+$/, '');
+  if (!trimmed) return '';
+  return trimmed.endsWith('/chat/completions') ? trimmed : `${trimmed}/chat/completions`;
+}
+
+function isLocalVirtualMacProvider(provider) {
+  return provider === LOCAL_VIRTUAL_MAC_PROVIDER || provider === 'local-virtual-mac-lane' || provider === 'qwen-mac-lane';
+}
+
+function xaiApiModel(model) {
+  return model === 'grok-4-3' ? 'grok-4.3' : model;
+}
+
+function isAnthropicClientModel(model) {
+  return /^claude(?:-|$)/i.test(coerceString(model)) || /^anthropic\//i.test(coerceString(model));
+}
+
+function providerRequiresApiKey(provider) {
+  return provider !== 'ollama' && !isLocalVirtualMacProvider(provider);
+}
+
 function loadRuntimeProfiles(local) {
   const profiles = local?.runtimeProfiles && typeof local.runtimeProfiles === 'object' ? local.runtimeProfiles : {};
   return {
@@ -53,9 +98,14 @@ function loadRuntimeProfiles(local) {
     xaiFallbackModel: coerceString(profiles.xaiFallbackModel || local?.xaiFallbackModel || local?.model?.xaiFallbackModel) || DEFAULT_XAI_FALLBACK_MODEL,
     nimFallbackModel: coerceString(profiles.nimFallbackModel || local?.nimFallbackModel || local?.model?.nimFallbackModel) || DEFAULT_NIM_FALLBACK_MODEL,
     xaiApiKey: coerceString(profiles.xaiApiKey || local?.xaiApiKey),
+    deepseekApiKey: coerceString(profiles.deepseekApiKey || local?.deepseekApiKey),
     nimApiKey: coerceString(profiles.nimApiKey || local?.nimApiKey),
     xaiBaseUrl: coerceString(profiles.xaiBaseUrl || local?.xaiBaseUrl) || DEFAULT_XAI_BASE_URL,
+    deepseekBaseUrl: coerceString(profiles.deepseekBaseUrl || local?.deepseekBaseUrl) || DEFAULT_DEEPSEEK_BASE_URL,
     nimBaseUrl: coerceString(profiles.nimBaseUrl || local?.nimBaseUrl) || DEFAULT_NIM_BASE_URL,
+    localFallbackModel: coerceString(profiles.localFallbackModel || local?.localFallbackModel) || DEFAULT_LOCAL_FALLBACK_MODEL,
+    localFallbackBaseUrl: normalizeChatCompletionsUrl(profiles.localFallbackBaseUrl || local?.localFallbackBaseUrl) || DEFAULT_LOCAL_FALLBACK_BASE_URL,
+    localFallbackApiKey: coerceString(profiles.localFallbackApiKey || local?.localFallbackApiKey),
   };
 }
 
@@ -104,7 +154,11 @@ function resolveProviderApiKey(provider, overrideApiKey = '', localApiKey = '')
     case 'openrouter':
       return process.env.OPENROUTER_API_KEY || '';
     case 'xai':
-      return process.env.XAI_API_KEY || '';
+      return process.env.XAI_API_KEY || process.env.GROK_API_KEY || process.env.ARIA_XAI_KEY || '';
+    case LOCAL_VIRTUAL_MAC_PROVIDER:
+    case 'local-virtual-mac-lane':
+    case 'qwen-mac-lane':
+      return process.env.ARIA_LOCAL_FALLBACK_API_KEY || process.env.ARIA_LANE_GATEWAY_API_KEY || '';
     case 'nim':
     case 'nvidia-nim':
       return process.env.NVIDIA_NIM_API_KEY || process.env.NIM_API_KEY || '';
@@ -116,16 +170,30 @@ function resolveProviderApiKey(provider, overrideApiKey = '', localApiKey = '')
 function resolveDefaultModel(provider, body, localModel, runtimeProfiles) {
   const explicitModel = coerceString(body?.llm?.model || body?.providerConfig?.model || body?.model);
   if (explicitModel) return explicitModel;
-  if (provider === 'deepseek') {
-    return isDeepArchitectureIntent(body) ? runtimeProfiles.deepModel : runtimeProfiles.chatModel;
-  }
-  if (provider === 'xai') return runtimeProfiles.xaiFallbackModel;
+  if (provider === 'deepseek') return runtimeProfiles.deepModel;
+  if (provider === 'xai') return runtimeProfiles.chatModel || runtimeProfiles.xaiFallbackModel;
   if (provider === 'nim' || provider === 'nvidia-nim') return runtimeProfiles.nimFallbackModel;
-  if (coerceString(localModel?.model)) return coerceString(localModel.model);
+  if (isLocalVirtualMacProvider(provider)) return runtimeProfiles.localFallbackModel;
+  if (localModel?.provider === provider && coerceString(localModel?.model)) return coerceString(localModel.model);
   if (provider === 'ollama') return DEFAULT_OLLAMA_MODEL;
   return '';
 }
 
+function resolveProviderNativeModel(provider, body, localModel, runtimeProfiles, fallbackModel = '') {
+  const overrideModel = coerceString(body?.llm?.model || body?.providerConfig?.model);
+  const clientModel = coerceString(body?.model);
+  if (overrideModel && (!clientModel || overrideModel !== clientModel || !isAnthropicClientModel(overrideModel))) {
+    return overrideModel;
+  }
+  if (localModel?.provider === provider && coerceString(localModel?.model)) return coerceString(localModel.model);
+  if (provider === 'deepseek') return runtimeProfiles.deepModel;
+  if (provider === 'xai') return runtimeProfiles.chatModel || runtimeProfiles.xaiFallbackModel;
+  if (provider === 'nim' || provider === 'nvidia-nim') return runtimeProfiles.nimFallbackModel;
+  if (isLocalVirtualMacProvider(provider)) return runtimeProfiles.localFallbackModel;
+  if (provider === 'ollama') return DEFAULT_OLLAMA_MODEL;
+  return isAnthropicClientModel(fallbackModel) ? '' : coerceString(fallbackModel);
+}
+
 function buildFallbackChain(body, runtimeProfiles) {
   if (body.allowProviderFallback === false) return [];
   const explicitFallbacks = Array.isArray(body?.llm?.fallbacks)
@@ -136,24 +204,24 @@ function buildFallbackChain(body, runtimeProfiles) {
   if (explicitFallbacks.length > 0) return explicitFallbacks;
   return [
     {
-      provider: 'xai',
-      model: runtimeProfiles.xaiFallbackModel,
-      baseUrl: runtimeProfiles.xaiBaseUrl,
-      apiKey: runtimeProfiles.xaiApiKey || process.env.XAI_API_KEY || '',
-      reason: 'xai reasoning fallback',
+      provider: 'deepseek',
+      model: runtimeProfiles.deepModel,
+      baseUrl: runtimeProfiles.deepseekBaseUrl,
+      apiKey: runtimeProfiles.deepseekApiKey || secretEnv('DEEPSEEK_API_KEY', 'ARIA_DEEPSEEK_API_KEY'),
+      reason: 'deepseek-v4-pro fallback',
     },
     {
-      provider: 'nim',
-      model: runtimeProfiles.nimFallbackModel,
-      baseUrl: runtimeProfiles.nimBaseUrl,
-      apiKey: runtimeProfiles.nimApiKey || process.env.NVIDIA_NIM_API_KEY || process.env.NIM_API_KEY || '',
-      reason: 'nim qwen fallback',
+      provider: LOCAL_VIRTUAL_MAC_PROVIDER,
+      model: runtimeProfiles.localFallbackModel,
+      baseUrl: runtimeProfiles.localFallbackBaseUrl,
+      apiKey: runtimeProfiles.localFallbackApiKey || secretEnv('ARIA_LOCAL_FALLBACK_API_KEY', 'ARIA_LANE_GATEWAY_API_KEY'),
+      reason: 'aria local virtual mac lane fallback',
     },
-  ].filter((entry) => entry.apiKey && entry.model);
+  ].filter((entry) => entry.model && (!providerRequiresApiKey(entry.provider) || entry.apiKey));
 }
 
 async function callResolvedProvider(config, body, ariaSystemPrompt) {
-  const model = body.model || config.model || 'gpt-4o';
+  const model = config.model || body.model || 'gpt-4o';
   const messages = composeOpenAIMessages(Array.isArray(body.messages) ? body.messages : [], ariaSystemPrompt);
   const maxTokens = Number(body.max_tokens || body.max_completion_tokens || 4096);
 
@@ -235,16 +303,28 @@ async function callResolvedProvider(config, body, ariaSystemPrompt) {
     xai: config.baseUrl || DEFAULT_XAI_BASE_URL,
     nim: config.baseUrl || DEFAULT_NIM_BASE_URL,
     'nvidia-nim': config.baseUrl || DEFAULT_NIM_BASE_URL,
+    [LOCAL_VIRTUAL_MAC_PROVIDER]: config.baseUrl || DEFAULT_LOCAL_FALLBACK_BASE_URL,
+    'local-virtual-mac-lane': config.baseUrl || DEFAULT_LOCAL_FALLBACK_BASE_URL,
+    'qwen-mac-lane': config.baseUrl || DEFAULT_LOCAL_FALLBACK_BASE_URL,
   };
   const providerUrl = urlByProvider[config.provider] || urlByProvider.openai;
+  const providerModel = config.provider === 'xai' ? xaiApiModel(model) : model;
   const extraHeaders = config.provider === 'openrouter'
     ? {
         'HTTP-Referer': 'http://127.0.0.1',
         'X-Title': 'Aria Mounted Runtime',
       }
+    : isLocalVirtualMacProvider(config.provider)
+      ? {
+          'X-Aria-Surface': body?.aria_surface || body?.metadata?.surface || 'client_chat',
+          'X-Aria-Role-Profile': body?.metadata?.roleProfile || 'interactive_chat',
+          'X-Aria-Stage': body?.metadata?.stage || 'fallback',
+          'X-Aria-Actor': body?.metadata?.actor || 'aria-smart-cli',
+          'X-Aria-System': body?.metadata?.system || 'aria-connector',
+        }
     : {};
   const data = await callOpenAICompatible(providerUrl, config.apiKey, {
-    model,
+    model: providerModel,
     messages,
     max_tokens: maxTokens,
     stream: false,
@@ -274,25 +354,33 @@ export function resolveProviderConfig(body = {}) {
     body.provider ||
     localModel?.provider ||
     coerceString(local?.defaultProvider) ||
-    (process.env.DEEPSEEK_API_KEY ? 'deepseek' : 'ollama');
+    (runtimeProfiles.xaiApiKey || secretEnv('XAI_API_KEY', 'GROK_API_KEY', 'ARIA_XAI_KEY')
+      ? 'xai'
+      : runtimeProfiles.deepseekApiKey || secretEnv('DEEPSEEK_API_KEY', 'ARIA_DEEPSEEK_API_KEY')
+        ? 'deepseek'
+        : 'ollama');
+  const localApiKey = localModel?.provider === provider ? coerceString(localModel?.apiKey) : '';
+  const localBaseUrl = localModel?.provider === provider ? coerceString(localModel?.baseUrl) : '';
   const merged = {
     provider,
     model: resolveDefaultModel(provider, body, localModel, runtimeProfiles),
     apiKey: resolveProviderApiKey(
       provider,
       coerceString(override?.apiKey || body.providerApiKey),
-      coerceString(localModel?.apiKey),
+      localApiKey,
     ),
     baseUrl:
       override?.baseUrl ||
       body.providerBaseUrl ||
-      localModel?.baseUrl ||
+      localBaseUrl ||
       (provider === 'ollama'
         ? DEFAULT_OLLAMA_BASE_URL
         : provider === 'deepseek'
-          ? DEFAULT_DEEPSEEK_BASE_URL
+          ? runtimeProfiles.deepseekBaseUrl
           : provider === 'xai'
-            ? DEFAULT_XAI_BASE_URL
+            ? runtimeProfiles.xaiBaseUrl
+            : isLocalVirtualMacProvider(provider)
+              ? runtimeProfiles.localFallbackBaseUrl
             : provider === 'nim' || provider === 'nvidia-nim'
               ? DEFAULT_NIM_BASE_URL
           : ''),
@@ -302,7 +390,7 @@ export function resolveProviderConfig(body = {}) {
     throw new Error('No provider configured. Save one in ~/.aria/config.json or pass llm/providerConfig in the request body.');
   }
 
-  if (merged.provider !== 'ollama' && merged.provider !== 'google' && !merged.apiKey) {
+  if (providerRequiresApiKey(merged.provider) && merged.provider !== 'google' && !merged.apiKey) {
     throw new Error(`No API key configured for provider ${merged.provider}. Save one in ~/.aria/config.json or pass llm.apiKey.`);
   }
 
@@ -415,7 +503,7 @@ async function callOpenAICompatible(url, apiKey, payload, extraHeaders = {}) {
       method: 'POST',
       headers: {
         'Content-Type': 'application/json',
-        Authorization: `Bearer ${apiKey}`,
+        ...(apiKey ? { Authorization: `Bearer ${apiKey}` } : {}),
         ...extraHeaders,
       },
       body: JSON.stringify(payload),
@@ -537,8 +625,11 @@ export async function callProviderForOpenAI(body, ariaSystemPrompt) {
 }
 
 export async function callProviderForAnthropic(body, ariaSystemPrompt) {
+  const local = loadLocalConfig();
+  const localModel = local?.model && typeof local.model === 'object' ? local.model : null;
+  const runtimeProfiles = loadRuntimeProfiles(local);
   const config = resolveProviderConfig(body);
-  const model = body.model || config.model || 'claude-sonnet-4-20250514';
+  const model = coerceString(body.model) || config.model || 'claude-sonnet-4-20250514';
   const req = composeAnthropicRequest(body, ariaSystemPrompt);
   const maxTokens = Number(body.max_tokens || 4096);
 
@@ -569,13 +660,25 @@ export async function callProviderForAnthropic(body, ariaSystemPrompt) {
       content: flattenAnthropicContent(message.content),
     })),
   ];
-  return callProviderForOpenAI({
+  const providerModel = resolveProviderNativeModel(config.provider, body, localModel, runtimeProfiles, config.model);
+  if (!providerModel) {
+    throw new Error(`No native model configured for provider ${config.provider}; refusing to send Anthropic client model ${model} upstream.`);
+  }
+  const providerMeta = await callProviderForOpenAI({
     ...body,
     provider: config.provider,
-    model,
+    model: providerModel,
     messages: openAiMessages,
-    llm: config,
+    llm: {
+      ...config,
+      model: providerModel,
+    },
   }, req.system);
+  return {
+    ...providerMeta,
+    model,
+    upstreamModel: providerMeta.model,
+  };
 }
 
 export function chunkText(text, maxChunkChars = 120) {