utilitas 1998.2.42 → 1998.2.43

package/lib/alan.mjs

@@ -628,8 +628,10 @@ const buildClaudeMessage = (text, options) => {
         }
     });
     return String.isString(text) ? {
-        role: options?.role || user,
-        content: [...attachments, { type: TEXT, text }],
+        role: options?.role || user, content: [...attachments, {
+            type: TEXT, text,
+            cache_control: options?.cache_control ?? { type: 'ephemeral' }
+        }],
     } : text;
 };
 
@@ -655,7 +657,7 @@ const streamResp = async (resp, options) => {
 };
 
 const getInfoEnd = text => Math.max(...[THINK_END, TOOLS_END].map(x => {
-    const keyEnd = text.indexOf(`${x}\n`);
+    const keyEnd = text.indexOf(text.endsWith(x) ? x : `${x}\n`);
     return keyEnd >= 0 ? (keyEnd + x.length) : 0;
 }));
 
@@ -1018,18 +1020,25 @@ const promptOllama = async (content, options = {}) => {
 
 const promptClaude = async (content, options = {}) => {
     options.model = options.model || DEFAULT_MODELS[CLAUDE];
-    let [_MODEL, event, text, thinking, signature, result, thinkEnd, tool_use]
-        = [MODELS[options.model], null, '', '', '', options.result ?? '', '', []];
+    let [
+        _MODEL, event, text, thinking, signature, result, thinkEnd, tool_use,
+        responded
+    ] = [
+            MODELS[options.model], null, '', '', '', options.result ?? '', '',
+            [], false
+        ];
     const { client } = await getClaudeClient(options);
     const { history }
         = await buildPrompts(_MODEL, content, { ...options, flavor: CLAUDE });
-    const resp = await client.messages.create({
+    const resp = await client.beta.messages.create({
         model: options.model, max_tokens: _MODEL.maxOutputTokens, ...history,
         stream: true, ...options.reasoning ?? _MODEL?.reasoning ? {
             thinking: options.thinking || { type: 'enabled', budget_tokens: 1024 },
         } : {}, ..._MODEL?.tools ? {  // https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking
             tools: options.tools ?? toolsClaude.map(x => x.def),
             tool_choice: { type: 'auto' },
+            betas: ['token-efficient-tools-2025-02-19'],
+            // @todo: https://docs.anthropic.com/en/docs/build-with-claude/tool-use/token-efficient-tool-use
         } : {},
     });
     for await (const chunk of resp) {
@@ -1048,6 +1057,8 @@ const promptClaude = async (content, options = {}) => {
             tool_use[tool_use.length - 1].input += event.partial_json;
         }
         deltaText = deltaThink + deltaText;
+        options.result && deltaText
+            && (responded = responded || (deltaText = `\n\n${deltaText}`));
         result += deltaText;
         deltaText && await streamResp({
             text: options.delta ? deltaText : result,
@@ -1143,7 +1154,6 @@ const promptGemini = async (content, options = {}) => {
         ) : {},
     });
     // https://github.com/google/generative-ai-js/blob/main/samples/node/advanced-chat.js
-    // @todo: check this issue similar to Vertex AI:
     // Google's bug: history is not allowed while using inline_data?
     const chat = client.startChat({ history, ...generationConfig(options) });
     const resp = await chat.sendMessageStream(prompt);

package/lib/manifest.mjs

@@ -1,7 +1,7 @@
 const manifest = {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.42",
+    "version": "1998.2.43",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",

package/package.json

@@ -1,7 +1,7 @@
 {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.42",
+    "version": "1998.2.43",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",