blun-king-cli 9.1.355 → 9.1.356

package/bin/system-prompt-token-cache-policy.cjs

@@ -0,0 +1,60 @@
+'use strict';
+
+const DEFAULT_MAX_ENTRIES = 4;
+const DEFAULT_MAX_PROMPT_CHARS = 500_000;
+const DEFAULT_MAX_TOTAL_CHARS = 1_000_000;
+
+function positiveInteger(value, fallback, name) {
+  const resolved = value === undefined ? fallback : value;
+  if (!Number.isSafeInteger(resolved) || resolved < 1) {
+    throw new TypeError(`${name} must be a positive integer`);
+  }
+  return resolved;
+}
+
+function createSystemPromptTokenEstimator(estimateTextTokens, options = {}) {
+  if (typeof estimateTextTokens !== 'function') {
+    throw new TypeError('estimateTextTokens must be a function');
+  }
+
+  const maxEntries = positiveInteger(options.maxEntries, DEFAULT_MAX_ENTRIES, 'maxEntries');
+  const maxPromptChars = positiveInteger(
+    options.maxPromptChars,
+    DEFAULT_MAX_PROMPT_CHARS,
+    'maxPromptChars',
+  );
+  const maxTotalChars = positiveInteger(
+    options.maxTotalChars,
+    DEFAULT_MAX_TOTAL_CHARS,
+    'maxTotalChars',
+  );
+  const cache = new Map();
+  let cachedChars = 0;
+
+  return function estimateSystemPrompt(prompt) {
+    const cached = cache.get(prompt);
+    if (cached !== undefined) {
+      cache.delete(prompt);
+      cache.set(prompt, cached);
+      return cached;
+    }
+
+    const tokens = estimateTextTokens(prompt);
+    if (prompt.length > maxPromptChars || prompt.length > maxTotalChars) return tokens;
+
+    while (cache.size >= maxEntries || cachedChars + prompt.length > maxTotalChars) {
+      const oldestPrompt = cache.keys().next().value;
+      if (oldestPrompt === undefined) break;
+      cachedChars -= oldestPrompt.length;
+      cache.delete(oldestPrompt);
+    }
+
+    cache.set(prompt, tokens);
+    cachedChars += prompt.length;
+    return tokens;
+  };
+}
+
+module.exports = {
+  createSystemPromptTokenEstimator,
+};

package/blun.mjs

@@ -31019,11 +31019,13 @@ function estimateTokensForContentPart(part) {
 		default: return 0;
 	}
 }
-var createToolSchemaTokenEstimator, estimateTokensForTools, messageTokenEstimateCache, MEDIA_TOKEN_ESTIMATE, MAX_IMAGE_HEADER_BYTES;
+var createToolSchemaTokenEstimator, createSystemPromptTokenEstimator, estimateTokensForTools, estimateSystemPromptTokens, messageTokenEstimateCache, MEDIA_TOKEN_ESTIMATE, MAX_IMAGE_HEADER_BYTES;
 var init_tokens = __esmMin((() => {
 	init_file_type$1();
 	({ createToolSchemaTokenEstimator } = createRequire(import.meta.url)("./bin/tool-schema-token-cache-policy.cjs"));
+	({ createSystemPromptTokenEstimator } = createRequire(import.meta.url)("./bin/system-prompt-token-cache-policy.cjs"));
 	estimateTokensForTools = createToolSchemaTokenEstimator(estimateTokens$1);
+	estimateSystemPromptTokens = createSystemPromptTokenEstimator(estimateTokens$1);
 	messageTokenEstimateCache = /* @__PURE__ */ new WeakMap();
 	MEDIA_TOKEN_ESTIMATE = 2e3;
 	MAX_IMAGE_HEADER_BYTES = 1024 * 1024;
@@ -74736,7 +74738,7 @@ var init_kosong_llm = __esmMin((() => {
 				const outgoingMessages = downgradeUnsupportedMedia(enrichedMessages, this.capability, (dropped) => {
 					this.notifyMediaDropped(dropped);
 				});
-				const outgoingRequestTokens = estimateTokens$1(this.systemPrompt) + estimateTokensForTools(tools) + estimateTokensForMessages(outgoingMessages);
+				const outgoingRequestTokens = estimateSystemPromptTokens(this.systemPrompt) + estimateTokensForTools(tools) + estimateTokensForMessages(outgoingMessages);
 				const reportedContextTokens = this.reportedContextTokens?.() ?? 0;
 				const usedContextTokens = Math.max(outgoingRequestTokens, reportedContextTokens);
 				completionBudget = applyCompletionBudgetWithDetails({
@@ -75454,7 +75456,7 @@ var init_full = __esmMin((() => {
 			return this.agent.context.tokenCountWithPending;
 		}
 		estimateRequestTokens(messages, systemPrompt = this.agent.effectiveSystemPrompt, tools = this.agent.tools.loopTools) {
-			return estimateTokens$1(systemPrompt) + estimateTokensForTools(tools) + estimateTokensForMessages(messages);
+			return estimateSystemPromptTokens(systemPrompt) + estimateTokensForTools(tools) + estimateTokensForMessages(messages);
 		}
 		resetForTurn() {
 			this.compactionCountInTurn = 0;
@@ -265521,7 +265523,7 @@ var init_agent = __esmMin((() => {
 						projectedTokenCount: this.fullCompaction.estimateCurrentRequestTokens(),
 						effectiveMaxContextTokens: this.fullCompaction.getEffectiveMaxContextTokens(),
 						compactionBudgetTokens: this.fullCompaction.getCompactionBudgetTokens(),
-						systemPromptTokens: estimateTokens$1(systemPrompt),
+						systemPromptTokens: estimateSystemPromptTokens(systemPrompt),
 						personalityPromptTokens,
 						socialPromptTokens,
 						skillPromptTokens: includedTokens(skillPrompt),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "blun-king-cli",
-  "version": "9.1.355",
+  "version": "9.1.356",
   "description": "BLUN CLI - your own AI agent with a Telegram channel. Get it done. With BLUN.",
   "license": "MIT",
   "bin": {