@peopl-health/nexus 6.0.0-dev.634 → 6.0.0-dev.649

package/lib/clinical/models/turnTraceModel.js

@@ -66,17 +66,22 @@ function buildIterationRecord(response, durationMs, apiCallConfig = {}) {
   const usage = response?.usage || {};
   const inputDetails = usage.input_tokens_details || {};
   const output = response?.output || [];
+  const content = Array.isArray(response?.content) ? response.content : [];
   return {
     model: apiCallConfig?.model || null,
     temperature: apiCallConfig?.temperature ?? null,
-    maxOutputTokens: apiCallConfig?.max_output_tokens || null,
+    maxOutputTokens: apiCallConfig?.max_output_tokens || apiCallConfig?.max_tokens || null,
     stopReason: response?.stop_reason || response?.status || null,
     inputTokens: usage.input_tokens || 0,
     outputTokens: usage.output_tokens || 0,
-    cacheCreationTokens: usage.cache_creation_tokens || inputDetails.cache_creation_tokens || 0,
-    cacheReadTokens: usage.cache_read_tokens || inputDetails.cached_tokens || 0,
-    reasoning: output.find((item) => item.type === 'reasoning')?.summary || null,
-    toolCallIds: output.filter((item) => item.type === 'function_call').map((item) => item.call_id),
+    cacheCreationTokens: usage.cache_creation_tokens || usage.cache_creation_input_tokens || inputDetails.cache_creation_tokens || 0,
+    cacheReadTokens: usage.cache_read_tokens || usage.cache_read_input_tokens || inputDetails.cached_tokens || 0,
+    reasoning: output.find((item) => item.type === 'reasoning')?.summary
+      || content.find((block) => block.type === 'thinking')?.thinking || null,
+    toolCallIds: [
+      ...output.filter((item) => item.type === 'function_call').map((item) => item.call_id),
+      ...content.filter((block) => block.type === 'tool_use').map((block) => block.id),
+    ].filter((id) => typeof id === 'string' && id.length > 0),
     responseId: response?.id || null,
     durationMs,
   };

package/lib/clinical/providers/AnthropicProvider.js

@@ -5,10 +5,14 @@ const { withTiming } = require('../../utils/tracingDecorator');
 
 const { logger } = require('../../utils/logger');
 
+const runtimeConfig = require('../../config/runtimeConfig');
+
 const { BaseLLMProvider } = require('./BaseLLMProvider');
 const { handleFunctionCalls } = require('./OpenAIResponsesProviderTools');
 
 const DEFAULT_MAX_FUNCTION_ROUNDS = 5;
+const CACHE_TTLS = ['5m', '1h'];
+const DEFAULT_CACHE_TTL = '1h';
 
 class AnthropicProvider extends BaseLLMProvider {
   constructor(options = {}) {
@@ -21,6 +25,11 @@ class AnthropicProvider extends BaseLLMProvider {
 
     this.client = client || new Anthropic({ apiKey });
     this.maxFunctionRounds = maxFunctionRounds;
+    const cacheTtl = runtimeConfig.get('ANTHROPIC_CACHE_TTL', DEFAULT_CACHE_TTL);
+    if (!CACHE_TTLS.includes(cacheTtl)) {
+      logger.warn('[AnthropicProvider] Invalid ANTHROPIC_CACHE_TTL, falling back to default', { cacheTtl, default: DEFAULT_CACHE_TTL });
+    }
+    this.cacheTtl = CACHE_TTLS.includes(cacheTtl) ? cacheTtl : DEFAULT_CACHE_TTL;
     this.defaults = {
       maxOutputTokens: 400,
       ...defaultModels,
@@ -41,13 +50,15 @@ class AnthropicProvider extends BaseLLMProvider {
     const accumulatedUsage = { input_tokens: 0, output_tokens: 0, total_tokens: 0 };
     const addUsage = (usage) => {
       if (!usage) return;
-      accumulatedUsage.input_tokens += usage.input_tokens || 0;
+      const inputTokens = (usage.input_tokens || 0) + (usage.cache_read_input_tokens || 0) + (usage.cache_creation_input_tokens || 0);
+      accumulatedUsage.input_tokens += inputTokens;
       accumulatedUsage.output_tokens += usage.output_tokens || 0;
-      accumulatedUsage.total_tokens += (usage.input_tokens || 0) + (usage.output_tokens || 0);
+      accumulatedUsage.total_tokens += inputTokens + (usage.output_tokens || 0);
     };
 
     const { system, messages } = this._toAnthropicMessages(instructions, input);
-    const apiCallConfig = { ...this._mapModelConfig(modelConfig), system };
+    const apiCallConfig = { ...this._mapModelConfig(modelConfig) };
+    if (system) apiCallConfig.system = system;
     if (toolSchemas.length > 0) {
       apiCallConfig.tools = this._mapTools(toolSchemas);
       apiCallConfig.tool_choice = this._mapToolChoice(toolChoice);
@@ -116,7 +127,11 @@ class AnthropicProvider extends BaseLLMProvider {
       messages.push({ role: item.role === 'assistant' ? 'assistant' : 'user', content });
     }
 
-    return { system: systemParts.join('\n\n'), messages };
+    const system = systemParts.length
+      ? [{ type: 'text', text: systemParts.join('\n\n'), cache_control: { type: 'ephemeral', ttl: this.cacheTtl } }]
+      : null;
+
+    return { system, messages };
   }
 
   _mapModelConfig(modelConfig) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@peopl-health/nexus",
-  "version": "6.0.0-dev.634",
+  "version": "6.0.0-dev.649",
   "description": "Core messaging and assistant library for WhatsApp communication platforms",
   "keywords": [
     "whatsapp",