@kunal_dutt/token-lens 0.2.22 → 0.2.25

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kunal_dutt/token-lens",
-  "version": "0.2.22",
+  "version": "0.2.25",
   "description": "Count and cut the tokens your prompts, chats and agents use, adapted to the model and task. MCP server and self-hosted gateway. Collects no data.",
   "keywords": [
     "mcp",

package/sdk.mjs

@@ -1,6 +1,16 @@
 // src/sdk/index.ts
 import { createHash as createHash4, randomUUID } from "node:crypto";
 
+// src/core/budget.ts
+var LONG_CHAT_TOKENS = 4e4;
+function contextUsage(conversationTokens, promptTokens, profile, windowOverride) {
+  const window = windowOverride && windowOverride > 0 ? windowOverride : profile.contextWindow;
+  const ratio2 = (conversationTokens + promptTokens) / window;
+  const zone = ratio2 >= 0.95 ? "full" : ratio2 >= 0.8 ? "hot" : ratio2 >= 0.5 ? "warm" : "ok";
+  return { conversationTokens, promptTokens, window, ratio: ratio2, zone, longChat: conversationTokens >= LONG_CHAT_TOKENS };
+}
+var HANDOFF_CAVEAT = "Optional: the new chat only knows what the brief keeps, so later answers can differ. Check the brief before you rely on it.";
+
 // src/core/advice.ts
 var fmt = (n) => n.toLocaleString("en-US");
 function advise(input) {
@@ -36,13 +46,13 @@ function advise(input) {
   if (usage.zone === "hot" || usage.zone === "full") {
     tips.push({
       id: "handoff",
-      text: `This chat fills about ${Math.round(usage.ratio * 100)}% of ${profile.label}'s context. Older messages will be dropped or compacted soon. Hand off to a new chat.`,
+      text: `This chat fills about ${Math.round(usage.ratio * 100)}% of ${profile.label}'s context. Older messages will be dropped or compacted soon. Hand off to a new chat. ${HANDOFF_CAVEAT}`,
       saves: "input"
     });
   } else if (usage.longChat || usage.zone === "warm") {
     tips.push({
       id: "handoff",
-      text: `Every new message re-reads about ${fmt(usage.conversationTokens)} tokens of history. A handoff brief and a new chat cut that.`,
+      text: `Every new message re-reads about ${fmt(usage.conversationTokens)} tokens of history. A handoff brief and a new chat cut that. ${HANDOFF_CAVEAT}`,
       saves: "input"
     });
   }
@@ -51,21 +61,12 @@ function advise(input) {
   if (lc && profile.pricing && reads >= lc.aboveTokens * 0.85)
     tips.push({
       id: "price-threshold",
-      text: reads > lc.aboveTokens ? `This message reads \u2248${fmt(reads)} tokens, past ${fmt(lc.aboveTokens)}, so ${profile.label} bills all of it at $${lc.inputPerMTok}/M instead of $${profile.pricing.inputPerMTok}/M. Continuing from a handoff brief in a new chat brings it back under.` : `This chat is \u2248${fmt(reads)} tokens; past ${fmt(lc.aboveTokens)}, ${profile.label} bills every message at $${lc.inputPerMTok}/M instead of $${profile.pricing.inputPerMTok}/M. Hand off to a new chat before then.`,
+      text: reads > lc.aboveTokens ? `This message reads \u2248${fmt(reads)} tokens, past ${fmt(lc.aboveTokens)}, so ${profile.label} bills all of it at $${lc.inputPerMTok}/M instead of $${profile.pricing.inputPerMTok}/M. Continuing from a handoff brief in a new chat brings it back under. ${HANDOFF_CAVEAT}` : `This chat is \u2248${fmt(reads)} tokens; past ${fmt(lc.aboveTokens)}, ${profile.label} bills every message at $${lc.inputPerMTok}/M instead of $${profile.pricing.inputPerMTok}/M. Hand off to a new chat before then. ${HANDOFF_CAVEAT}`,
       saves: "input"
     });
   return tips;
 }
 
-// src/core/budget.ts
-var LONG_CHAT_TOKENS = 4e4;
-function contextUsage(conversationTokens, promptTokens, profile, windowOverride) {
-  const window = windowOverride && windowOverride > 0 ? windowOverride : profile.contextWindow;
-  const ratio2 = (conversationTokens + promptTokens) / window;
-  const zone = ratio2 >= 0.95 ? "full" : ratio2 >= 0.8 ? "hot" : ratio2 >= 0.5 ? "warm" : "ok";
-  return { conversationTokens, promptTokens, window, ratio: ratio2, zone, longChat: conversationTokens >= LONG_CHAT_TOKENS };
-}
-
 // node_modules/gpt-tokenizer/esm/constants.js
 var DEFAULT_MERGE_CACHE_SIZE = 1e5;